SEMICONDUCTOR GAS SENSORS This Page Intentionally Left Blank Woodhead Publishing Series in Electronic and Optical Materials

SEMICONDUCTOR GAS SENSORS This page intentionally left blank Woodhead Publishing Series in Electronic and Optical Materials

SEMICONDUCTOR GAS SENSORS

Second Edition

Edited by RAIVO JAANISO University of Tartu, Tartu, Estonia OOI KIANG TAN Nanyang Technological University, Singapore Woodhead Publishing is an imprint of Elsevier The Ofﬁcers’ Mess Business Centre, Royston Road, Duxford, CB22 4QH, United Kingdom 50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States The Boulevard, Langford Lane, Kidlington, OX5 1GB, United Kingdom

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Details on how to seek permission, further information about the Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.

This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be noted herein).

Notices Knowledge and best practice in this ﬁeld are constantly changing. As new research and experience broaden our understanding, changes in research methods, professional practices, or medical treatment may become necessary.

Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility.

To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein.

Library of Congress Cataloging-in-Publication Data A catalog record for this book is available from the Library of Congress

British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library

ISBN: 978-0-08-102559-8

For information on all Woodhead Publishing publications visit our website at https://www.elsevier.com/books-and-journals

Publisher: Matthew Deans Acquisition Editor: Kayla Dos Santos Editorial Project Manager: Ali Afzal-Khan Production Project Manager: Debasish Ghosh Cover Designer: Matthew Limbert

Typeset by TNQ Technologies Contents

Contributors xi

Part One Basics

1. Fundamentals of semiconductor gas sensors 3 Noboru Yamazoe and Kengo Shimanoe 1.1 Introduction 4 1.2 Classiﬁcation of semiconductor gas sensors 5 1.3 Resistor-type sensors: empirical aspects 6 1.4 Resistor-type sensors: theoretical aspects 14 1.5 Future trends 34 References 37

2. Conduction mechanism in semiconducting metal oxide sensing ﬁlms: impact on transduction 39 N. B^arsan, M. Huebner and U. Weimar 2.1 Introduction 39 2.2 General discussion about sensing with semiconducting metal oxide gas sensors 41 2.3 Sensing and transduction for p- and n-type semiconducting metal oxides 47 2.4 Investigation of the conduction mechanism in semiconducting metal oxide sensing layers: studies in working conditions 57

2.5 Conduction mechanism switch for n-type SnO2–based sensors during operation in application-relevant conditions 66 2.6 Conclusion and future trends 67 References 67

3. The effect of electrode-oxide interfaces in gas sensor operation 71 Sung Pil Lee and Chowdhury Shaestagir 3.1 Introduction 72 3.2 Electrode materials for semiconductor gas sensors 74 3.3 Electrode-oxide semiconductor interfaces 95 3.4 Charge carrier transport in the electrode-oxide semiconductor interfaces 104

v j vi Contents

3.5 Gas/solid interactions in the electrode-oxide semiconductor interfaces 119 3.6 Conclusions 124 References 125

4. Introduction to semiconductor gas sensors: a block scheme description 133 Arnaldo D’Amico and Corrado Di Natale

4.1 Introduction 133 4.2 The sensor blocks 135 4.3 Metal oxide semiconductor capacitor: the case of the

hydrogen gas sensitivity of Pd-SiO2-Si 142 4.4 Light-addressable potentiometric sensor 144 4.5 Metal oxide semiconductor ﬁeld-effect transistor 148 4.6 Metal oxide semiconductors 151 4.7 Conclusions 156 References 156

Part Two Materials

5. One- and two-dimensional metal oxide nanostructures for chemical sensing 161 E. Comini and D. Zappa 5.1 Introduction 161 5.2 Deposition techniques 162 5.3 Conductometric sensor 169 5.4 Transduction principles and related novel devices 170 5.5 Conclusion and future trends 174 References 175

6. Hybrid materials with carbon nanotubes for gas sensing 185 Thara Seesaard, Teerakiat Kerdcharoen and Chatchawal Wongchoosuk

6.1 Introduction 186 6.2 Synthesis of carbon nanotube 192 6.3 Preparation of carbon nanotubedmetal oxide sensing ﬁlms 194 6.4 Sensor assembly 199 6.5 Characterization of carbon nanotube–metal oxide materials 200 6.6 Sensing mechanism of carbon nanotube–metal oxide gas sensors 205 Contents vii

6.7 Fabrication of electrodes and CNT/polymer nanocomposites for textile-based sensors 206 6.8 Sensor assembly for textile-based gas sensors 210 6.9 Characterization of CNT/polymer nanocomposites sensing materials on textile substrate 212 6.10 Sensing mechanism of CNT/polymer nanocomposites sensing materials on fabric substrate 215 6.11 Conclusion 216 Acknowledgments 217 References 217

7. Carbon nanomaterials functionalized with macrocyclic compounds for sensing vapors of aromatic VOCs 223 Pierrick Clément and Eduard Llobet

7.1 Introduction 223 7.2 Cyclodextrins 226 7.3 Calixarenes and derivatives 229 7.4 Deep cavitands 230 7.5 Conclusions 232 Acknowledgments 235 References 235

8. Luminescence probing of surface adsorption processes using InGaN/GaN nanowire heterostructure arrays 239 Konrad Maier, Andreas Helwig, Gerhard Muller€ and Martin Eickhoff 8.1 Adsorptiondkey to understanding semiconductor gas sensors 239 8.2 III-nitrides as an emerging semiconductor technology 243 8.3 Photoluminescent InGaN/GaN nanowire arrays 243 8.4 Optical probing of adsorption processes 245 8.5 Experimental observations of PL response 246 8.6 Analysis of adsorption phenomena 250 8.7 Molecular mechanism of adsorption 261 8.8 Conclusions and outlook 266 References 267

9. Rare earth–doped oxide materials for photoluminescence-based gas sensors 271 V. Kiisk and Raivo Jaaniso 9.1 Introduction 272 3þ 9.2 Sm :TiO2 277 3þ 9.3 Eu :ZrO2 288 viii Contents

3þ 9.4 Tb :CePO4 294 3þ 9.5 Pr :(K0.5Na0.5)NbO3 298 9.6 Conclusion 299 References 300

Part Three Methods and integration

10. Recent progress in silicon carbide ﬁeld effect gas sensors 309 M. Andersson, A. Lloyd Spetz and D. Puglisi

10.1 Introduction 309 10.2 Background: transduction and sensing mechanisms 312 10.3 Sensing layer development for improved selectivity of SiC gas sensors 327 10.4 Dynamic sensor operation and advanced data evaluation 332 10.5 Applications 335 10.6 Summary 338 Acknowledgments 217 References 339

11. Semiconducting direct thermoelectric gas sensors 347 F. Rettig and R. Moos

11.1 Introduction 347 11.2 Direct thermoelectric gas sensors 353 11.3 Conclusion and future trends 380 References 381

12. Dynamic operation of semiconductor sensors 385 Andreas Schutze€ and Tilman Sauerwald 12.1 Introduction 385 12.2 Dynamic operation of metal oxide semiconductor gas sensors 388 12.3 Dynamic operation of gas-sensitive ﬁeld-effect transistors 398 12.4 Conclusion and outlook 404 References 408

13. Micromachined semiconductor gas sensors 413 D. Briand and J. Courbat 13.1 Introduction 413 13.2 A brief history of semiconductors as gas-sensitive devices 414 13.3 Microhotplate concept and technologies 416 13.4 Micromachined metal oxide gas sensors 425 Contents ix

13.5 Complementary metal oxide semiconductor–compatible metal oxide gas sensors 437 13.6 Micromachined ﬁeld-effect gas sensors 442 13.7 Nanostructured gas sensing layers on microhotplates 445 13.8 Semiconductor gas sensors on polymeric foil and their additive manufacturing 450 13.9 Manufacturing, products, and applications 454 13.10 Conclusion 458 References 459

14. Integrated CMOS-based sensors for gas and odor detection 465 P.K. Guha, S. Santra and J.W. Gardner

14.1 Introduction 465 14.2 Microresistive complementary metal oxide semiconductor gas sensors 467 14.3 Microcalorimetric complementary metal oxide semiconductor gas sensor 469 14.4 Sensing materials and their deposition on complementary metal oxide semiconductor gas sensors 472 14.5 Interface circuitry and its integration 475 14.6 Integrated multisensor and sensor array systems 480 14.7 Conclusion and future trends 483 Useful web addresses 485 References 486

Index 489 This page intentionally left blank Contributors

M. Andersson Linkoping€ University, Linkoping,€ Sweden N. Bârsan University of Tubingen,€ Tubingen,€ Germany D. Briand Ecole Polytechnique Fédérale de Lausanne, Neuchâtel, Switzerland Pierrick Clément Microsystems Laboratory, Ecole Polytechnique Féderale de Lausanne (EPFL), Lausanne, Switzerland E. Comini Department of Information Engineering, University of Brescia, Brescia, Italy J. Courbat Formely Ecole Polytechnique Fédérale de Lausanne, Neuchâtel, Switzerland Arnaldo D’Amico Department of Electronic Engineering, University of Rome Tor Vergata, Roma, Italy Corrado Di Natale Department of Electronic Engineering, University of Rome Tor Vergata, Roma, Italy Martin Eickhoff Institute of Solid State Physics, University of Bremen, Bremen, Germany J.W. Gardner University of Warwick, Coventry, United Kingdom P.K. Guha Indian Institute of Technology, Kharagpur, West Bengal, India Andreas Helwig Airbus Group Innovations, Munich, Germany M. Huebner University of Tubingen,€ Tubingen,€ Germany Raivo Jaaniso University of Tartu, Tartu, Estonia Teerakiat Kerdcharoen Department of Physics and NANOTEC Center of Excellence, Faculty of Science, Mahidol University, Ratchathewi, Bangkok, Thailand V. Kiisk University of Tartu, Tartu, Estonia Sung Pil Lee Kyungnam University, Changwon, Kyungnam, Korea

xi j xii Integrated CMOS-based sensors for gas and odor detectionContributors

Eduard Llobet MINOS-EMaS, Department of Electronic Engineering, Universitat Rovira i Virgili, Tarragona, Spain A. Lloyd Spetz Linkoping€ University, Linkoping,€ Sweden Konrad Maier Airbus Group Innovations, Munich, Germany R. Moos University of Bayreuth, Bayreuth, Germany Gerhard Muller€ Department of Applied Sciences and Mechatronics, Munich University of Applied Sciences, Munich, Germany D. Puglisi Linkoping€ University, Linkoping,€ Sweden F. Rettig University of Bayreuth, Bayreuth, Germany S. Santra Indian Institute of Technology, Kharagpur, West Bengal, India Tilman Sauerwald Lab for Measurement Technology, Department Systems Engineering, Saarland University, Saarbrucken,€ Germany Andreas Schutze€ Lab for Measurement Technology, Department Systems Engineering, Saarland University, Saarbrucken,€ Germany Thara Seesaard Department of Physics, Faculty of Science and Technology, Kanchanaburi Rajabhat University, Muang District, Kanchanaburi, Thailand Chowdhury Shaestagir Intel Corporation, Hillsboro, OR, United States Kengo Shimanoe Faculty of Engineering Sciences, Kyushu University, Kasuga, Fukuoka, Japan U. Weimar University of Tubingen,€ Tubingen,€ Germany Chatchawal Wongchoosuk Department of Physics, Faculty of Science, Kasetsart University, Chatuchak, Bangkok, Thailand Noboru Yamazoe Faculty of Engineering Sciences, Kyushu University, Kasuga, Fukuoka, Japan D. Zappa Department of Information Engineering, University of Brescia, Brescia, Italy PART ONE

Basics

1j This page intentionally left blank CHAPTER ONE

Fundamentals of semiconductor gas sensors

Noboru Yamazoe, Kengo Shimanoe Faculty of Engineering Sciences, Kyushu University, Kasuga, Fukuoka, Japan

Contents

1.1 Introduction 4 1.2 Classification of semiconductor gas sensors 5 1.3 Resistor-type sensors: empirical aspects 6 1.3.1 Sensing materials and devices 6 1.3.1.1 Sensing materials 6 1.3.1.2 Sensitizers 8 1.3.1.3 Device structure 9 1.3.1.4 Fabrication 10 1.3.2 Gas sensing characteristics 11 1.3.2.1 Response and response transients 11 1.3.2.2 Operating temperature 12 1.3.2.3 Disturbances to gas response 13 1.3.3 Semiconductor oxygen sensors 13 1.4 Resistor-type sensors: theoretical aspects 14 1.4.1 Receptor function and transducer function 14 1.4.2 Response to oxygen (base air resistance) 18 1.4.3 Response to inflammable gases 22 1.4.4 Response to oxidizing gases 23 1.4.5 Extensions 25 1.4.6 Nonresistive sensors 27 1.4.7 Field-effect transistor-type gas sensors 27 1.4.7.1 Principle 27 1.4.7.2 Solid electrolyte-gate field-effect transistor 28 1.4.7.3 Oxide semiconductor-gate field-effect transistor 29 1.4.7.4 Dielectric material-gate field-effect transistor 31 1.4.8 Oxygen concentration cell type sensors 31 1.4.9 Other gas sensors 32 1.4.9.1 Metaleinsulatoresemiconductor capacitor type sensors 32 1.4.9.2 Diode-type sensors 33

1.5 Future trends 34 1.5.1 Needs and seeds 34 1.5.2 Basic approaches desired 35 1.5.3 Challenges 36 References 37

1.1 Introduction

Semiconductor gas sensors using metal oxides such as SnO2 were pioneered by two research groups in Japan.1,2 These sensors were soon put on the market as gas leak alarms and proved to be indispensable in keeping people safe from the distressing circumstances resulting from gas leaks. At the same time, their success had worldwide impact on researchers, creating awareness of the importance of gas sensors or chemical sensors more generally. Great effort has subsequently been made in the development of new gas sensors, including those using silicon semiconductor devices and solid electrolytes devices. If the definition of a semiconductor gas sensor is a sensor into which a semiconductor material is incorporated, there is a variety of semiconductor gas sensors of varying structures, made of different materials and involving various working principles. This introduction describes the fundamental aspects of the various semiconductor gas sensors that have been developed so far, or that are proposed. First, they are classified into five types, based on the constitutional principle of sensor devices (Section 1.2). The structure of devices, their working principles, and sensing mechanisms are described in subsequent sections. However, the greatest space is devoted to describing experimental knowledge and the theory of gas response of the sensors based on resistors, which have been made full use of and which still have potential for further development. It has long been queried why sensors of this type are promoted with regard to their sensitivity, as the constituent oxides are smaller than in other types of device,3 though a semiempirical analysis has been attempted.4,5 This issue was recently resolved by developing a new theory on the receptor function of small-sized oxides.6,7 As revealed in the new theory, small semiconductors are depleted of electrons in two stages by a process of ionosorption of oxygen or oxidizing gases, resulting in the appearance of regional depletion followed by volume depletion. Gas response can be sufficiently understood based on the same theory. It is shown that the theory gives an important clue to understanding the gas Fundamentals of semiconductor gas sensors 5 response of oxides attached to potentiometric gas sensors (Section 1.5). The chapter closes with personal observations regarding semiconductor gas sensors (Section 1.6).

1.2 Classification of semiconductor gas sensors Generally speaking, a gas sensor is composed of a receptor and a transducer, as illustrated in Fig. 1.1. The former is provided with a material or a materials system which, on interacting with a target gas, either induces a change in its own properties (work function, dielectric constant, electrode potential, mass, etc.) or emits heat or light. The transducer is a device to transform such an effect into an electrical signal (sensor response). The construction of a sensor is determined by the transducer used, with the receptor appearing to be implanted within it. From this perspective, a semiconductor gas sensor can be defined as a sensor in which a semiconductor material is used as a receptor and/or transducer. There are two groups of semiconductors: oxide and nonoxide (typically, silicon). Nonoxide semiconductors cannot work as a receptor because they are coated with a protective insulation layer, but they can provide a transducer in the form of MIS FETs (metaleinsulatoresemiconductor field- effect transistor) and MIS capacitors. In contrast, oxide semiconductors can work as both a receptor and a transducer (mostly in the form of a resistor)

Figure 1.1 Gas sensor as constituted of a receptor and a transducer. R ¼ resistance,

E ¼ electromotive force, I ¼ current, Vth ¼ threshold voltage (FET), Cp ¼ capacitance. 6 Noboru Yamazoe and Kengo Shimanoe owing to their chemical and physical stability in hostile environments at elevated temperatures. Table 1.1 shows various examples of semiconductor gas sensors classified according to the types of transducer used and subclassified by the kinds of receptor used, together with the kinds of signal output (response), typical sensor devices, and the gases targeted. The transducers are seen to be available in the forms of resistors, diodes, MIS capacitors, MIS FETs, or oxygen concentration cells. For each type of sensor thus classified, devices, sensing principles, and the important features of semiconductor gas sensors are now described.

1.3 Resistor-type sensors: empirical aspects Of the various types of sensor, resistor sensors have received the greatest investigation and have proven their feasibility in practice. These sensors are often called “oxide semiconductor gas sensors.” There are two subtypes: surface sensitive and bulk sensitive. This section is devoted to surface- sensitive resistor sensors, except for Section 1.3.3 which briefly discusses bulk sensitive resistor sensors. It is noted that books and review articles e have been published about oxide semiconductor gas sensors.8 10 1.3.1 Sensing materials and devices 1.3.1.1 Sensing materials A surface-sensitive resistor sensor works on a very simple principle; on exposure to a target gas in air at an elevated temperature, its resistance either decreases or increases as a function of the partial pressure of the gas. Of the many metal oxides, n-type oxides (SnO2,In2O3,WO3, ZnO, and g-Fe2O3) and p-type oxides (CuO and Co3O4) exhibit significant gas sensing properties. Mainly because of stability issues, however, SnO2,In2O3, and WO3 have been adopted as the sensor materials utilized in practice. In practice, even these oxides are frequently loaded or mixed empirically with several foreign materials as a sensitizer (PdO, Pt, Fe2O3, etc), a skeleton material (alumina), or a binder (silica). When an n-type oxide is used, resistance decreases on exposure to inflammable or reducing gases in the air (inorganic: H2, CO, NH3,H2S, NO, etc; organic: CH4, propane, alcohols, odorants, etc.), while it increases on exposure to oxidative gases (NO2, ozone, N2O, etc.). Apart from such redox-active gases, CO2 and water vapor have been known to affect the resistance to a greater or lesser degree. Exploitation of the effects of CO2 11 has led to the development of a semiconductor CO2 sensor. udmnaso eiodco a sensors gas semiconductor of Fundamentals

Table 1.1 Classiﬁcation of semiconductor gas sensors according to the types of transducers and receptors used. Transducer Response signal Receptor Device (example) Target

Resistor Resistance Oxides Porous SnO2 (surface- A variety of gases sensitive) Sintered TiO2 (bulk- Air/fuel ratio (car sensitive) engine) Diode Bias current Oxides Pd-TiO2 (single crystal) H2 Metaleinsulator Bias potential shift Pd Pd-gate capacitor H2,NH3 esemiconductor (MIS) capacitor MIS ﬁeld-effect Threshold voltage shift Pd Pd-gate FET H2,NH3 transistor (FET) Ionic Proton H2 conductors conductor gate FET NaNO2-gate FET NO2 Oxides WO3-gate FET NO2 Dielectrics Cellulose-gate FET Humidity Oxygen concentration Cell voltage Oxides Pt/zirconia/oxide/Pt A variety of gases cell

Note: “Oxides” stands for semiconductive metal oxides. 7 8 Noboru Yamazoe and Kengo Shimanoe

1.3.1.2 Sensitizers Gas sensing properties, especially gas responses, are known to be often improved signiﬁcantly when constituent oxides are loaded with small amounts of appropriately chosen foreign materials. Examples are SnO2- PdO (CO, propane, etc.), SnO2-Pt and/or PdO (methane), SnO2-Co3O4 (CO), SnO2-CuO (H2S), SnO2-Ag2O(H2), In2O3-PdO (CO, odorant gases), WO3-Au (NH3), SnO2-La2O3-Pt (ethanol), SnO2-CaO (ethanol), In2O3-Fe2O3 (ozone), SnO2-Fe2O3 (NO2), TiO2-Cr2O3 (NO), etc. In this list, the materials following the oxide semiconductors are sensitizers and the target gases are shown in parentheses. As suggested from the large variation in sensitizers, the mechanisms of sensitization involved are not so simple. It is useful to know that the dispersion of the sensitizers, except Pt, always causes the resistances of the device in base air to increase. This suggests that those interact with the oxides and increase the work function of the oxides. In view of heterogeneous catalysis, Pt, PdO, CuO, Ag2O, Co3O4, and Au are well-known oxidation catalysts to reducing gases. Therefore, such catalytic activity is relevant to the sensitizing actions. It should be noted, however, that the mere promotion of oxidation reactions cannot contribute to gas response unless it has something to do with the surface properties of the oxides. In this sense, the sensitizers, except Pt, undergo redox changes such as PdO þ H2 / Pd þ H2O, Pd þ (1/2) O2 / PdO, and the changes of their redox state on exposure to target gases can possibly induce changes in device resistance (gas response) through electronic interactions with oxides (electronic sensitization). In the case of Pt, on the other hand, it seems that the target gas (methane) is partially oxidized on Pt to HCHO or CO, which then reacts with the adsorbed oxygen of the oxide (chemical sensitization). La2O3 and CaO, which have no such catalytic oxidation activity, modify the acid-base properties of the oxide surface more basic; on the acidic surface, ethanol undergoes dehydration (no consumption of O ), C2H5OH / C2H4 þ H2O; on the basic surface, it undergoes oxidative dehydrogenation, C2H5OH þ 2O / C2H4O þ H2O. It is thus understood that, in this case, the selectivity of reaction paths is changed by the sensitizers. As shown above, Fe2O3 promotes response to oxidizing gases, though the mechanism of promotion is not yet clear. There can be no doubt that sensitizers are very important for practical devices. Unfortunately, however, little basic research has been carried out on sensitizers and sensitizing actions. Fundamentals of semiconductor gas sensors 9

1.3.1.3 Device structure Sensor devices are fabricated into a resistor in which a porous stack of the sensing materials is attached with a heater and a resistance measuring probe (usually a pair of metal electrodes). Various structures have been devised in practice, as shown in Fig. 1.2. Originally fabrication was a sintered block structure (about 0.5 cm in size) with a pair of Pt coil electrodes inserted (a); one of the coils also served as a heater. This was followed by a thin alumina tube within a heavy coating (b); a pair of wire electrodes was wound on the tube and a heater was set inside it. Currently in wide use is a thick ﬁlm structure (c), screen-printed on an alumina substrate with a pair of electrodes, and a heater printed in advance. A microversion of this structure,

Figure 1.2 Device structures adopted for resistor-type sensors in practice. (a) Sintered block, (b) thin alumina tube-coated layer, (c) screen printed thick film, (d) small bead inserted with coil and needle electrodes, (e) small bead inserted with a single coil (heater and electrode), (f) practical sensor element assembling sensor device, metal cap, and filter. 10 Noboru Yamazoe and Kengo Shimanoe known as a MEMS (microelectromechanical system) sensor, is currently under development, as will be described later. Apart from these standard structures, bead-shaped structures have been devised for practical use. A small bead made of sensing materials (about 0.5 mm in size) is inserted with a coil and needle electrodes in (d); the coil also works as a heater. A similar bead is inserted with a single coil (heater) in (e), the so-called “hot wire” type; a change in the resistance of the sensing materials affects the composite resistance between the two terminals of the inserted coil, which is measured precisely on a bridge circuit as gas response. For actual use, each device is bonded to the connector pins and put inside a metal cap with a hole(s) on top to remove the risk of triggering gas explosions. In addition, an adsorbent such as active carbon (often referred to as a “filter”) is placed in a layer immediately behind the hole to remove unpleasant gases, as shown in (f).

1.3.1.4 Fabrication Important guidelines for device fabrication collected empirically can be summarized as follows: 1. Crystallite sizes of oxide semiconductors should be as small as possible. 2. Sensitizers should be dispersed as finely as possible. 3. Sensing layer thickness and porosity should also be optimized to improve selectivity and durability. According to these guidelines, fabrication of devices is carried out carefully. It starts with the preparation of a fine powder of oxide semiconductor (crystallite size around 10 nm in diameter) through what is known as a “wet” process. This is the precipitation of a precursor of the oxide from an aqueous solution of its metal salt(s), followed by the gentle washing, drying, and calcination of the precursor before its conversion to the final powder. The powder is loaded with a small amount of a sensitizer and then converted into slurry (paste) by milling it using water or organic vehicles, together with any other necessary additives. The slurry is finally deposited over the electrodes (block or bead type) or on the substrate (thick film type), and, after drying, the deposit is sintered under specific conditions to stabilize the porous microstructure. It is noted that all of the semiconductor gas sensors so far in use are of the thick film (or layer) type, prepared through the wet processes discussed above. Thin film type devices, especially those fabricated via physical methods such as sputtering, have frequently shown interesting sensing performances in the short term, but little use is currently made of these devices. Fundamentals of semiconductor gas sensors 11

1.3.2 Gas sensing characteristics 1.3.2.1 Response and response transients The behavior of resistance on switching between base air and gas ambient is illustrated in Fig. 1.3(a). On switching to an inflammable gas ambient, the resistance reduces from a value in air (Ra) to a stationary value (Rg), while it goes back to Ra on switching back. Empirically, gas response is defined as the ratio Ra/Rg (normalized conductance). The rate of response or recovery is expressed empirically in terms of the time (s) needed for a 90% full response or recovery. In the case of oxidizing gases such as NO2, which increase the resistance, gas response is defined as Rg/Ra (normalized resistance). The dependence of Rg on the partial pressure of target gas (Pg)is known empirically to fall on linear correlations on logarithmic scales12; ¼ a that is, Rg cPg , where a and c are constants (power law), as shown in Fig. 1.3(b). Accordingly, gas response also follows power law, = ¼ a fl = ¼ a Ra Rg cPg (in ammable gases) or Rg Ra cPg (oxidizing gases).

Figure 1.3 Response and recovery transients. (a) On switching on and off an inﬂam- mable gas in air, (b) linear correlation observed between resistance (Rg) and partial pressure of the gas (Pg) on logarithmic scales (power law). 12 Noboru Yamazoe and Kengo Shimanoe

The power index, a, is almost fixed depending on the kinds of target gas, taking values roughly equal to 1/2 to many inflammable gases (H2, CO, etc), 1 to NO2, and 1/2 for O3. It is noted that the resistance under exposure to varying partial pressure of oxygen (PO2) follows the power equation with = a ¼ 1=2, namely, R ¼ c'''P1 2. The power indices are related to the O2 O2 modes of interaction between the gases and the surface of oxide semiconductors, as will be discussed later. Sensitivity is usually defined as a slope of the correlation between gas response and Pg. In the event that power law holds well, however, this definition is meaningless because sensitivity is dependent on Pg unless a fi a is unity. This dif culty is overcome if Pg is replaced by Pg in the above definition. The slope (sensitivity) is then nothing but the proportionality constant of the power equation. Sensitivity is determined by the physicochemical constants of semiconductor, target gas, and oxygen. 1.3.2.2 Operating temperature Response and response transients are sensitive to the operating temperature. The rates of response and recovery naturally increase with increasing temperature. On the other hand, response shows different behavior depending on whether the gas is inflammable or oxidizing. For an inflammable gas, response goes through a maximum on increasing temperature, resulting in a well-known bell-shaped correlation between the response and temperature. This dependence appears because the rate constant of the surface reaction between gas and adsorbed oxygen (kR) increases exponentially with a rise in temperature, while the Knudsen diffusion coefficient of the gas (DK) does so sublinearly. In the lower temperature region, kR < DK is held so that kR is an exclusive determinant for gas response. In higher temperatures, on the other hand, the relation is inversed, kR > DK, and the response is attenuated by the gas diffusion and reaction effect.13,14 In this temperature range, the gas is consumed significantly by diffusion from the surface to the inside of the porous sensing layer. The effective partial pressure of the gas in the inner region where the resistance is actually measured can be significantly lower than the nominal value outside. The ratio of the actual gas response to the ideal (free of attenuation) is known as the “utility factor” (U). U remains unity in lower temperatures, while in higher temperatures it decreases rather sharply with increasing temperature, increasing diffusion length (sensing layer thickness), and decreasing pore size. It follows that the response maximum and the temperature at that point vary not only by the kinds of gas and oxide semiconductor Fundamentals of semiconductor gas sensors 13 but also by the device structure (layer thickness, in particular) and the sensing materials adopted in processing. Strictly speaking, there is a further possible reason for the decrease of gas response at high temperature: oxygen adsorption is decreased with increasing temperature. Therefore, if the partial pressure of the inflammable gas is too great, adsorbed oxygen is consumed (resistance reaches minimum) such that gas response will decrease with increasing temperature, reflecting the temperature dependence of the adsorbed oxygen. This discussion is valid for a small partial pressure of gas. Oxidizing gases such as NO2, on the other hand, are adsorbed on the oxide semiconductor particles. The amount of adsorption, and therefore the gas response, increases as the temperature drops. Operating temperature is then determined as a compromise between gas response and rates of response and recovery.

1.3.2.3 Disturbances to gas response Gas response reacts to disturbances to varying degrees. There are two kinds of disturbance: a drift of base air resistance (Ra) and a modulation of gas response (Rg) by coexistent gases. As for the former, Ra shifts downward quickly on increasing the partial pressure of coexistent water vapor “ ” (PH2O), a phenomenon known as a short-term effect of water vapor.

Apart from this phenomenon, PH2O seems to be related to a long-term drift of Ra; it is known that Ra undergoes seasonal changes; that is, it goes up in summer and goes down in winter. Unfortunately, these two types of drifts are yet to be clarified in detail. Practically, attempts have been made to correct the long-term drift partly by means of software. The disturbance brought about by a modulation of gas response can be simplified if both the target gas and coexistent gas are inflammable, as in the case of sensing CO in the coexistence of H2. The strength of the disturbance can be estimated if the sensitivity to each gas is known. To mitigate interfer- ences by coexistent gases, nonstandard modes of sensor operation have been adopted in some cases for sensing CO and alcohol in the breath. 1.3.3 Semiconductor oxygen sensors At sufficiently high temperatures, where the bulk diffusion of component ion oxides is activated to a significant degree, oxide semiconductors are known to change nonstoichiometry, and thus electronic conductivity fl changes depending on PO2. On exposure to a mixture of in ammable gas and air, sensors using such oxides change resistance depending on the composition of the mixture. What is responsible for the change in resistance 14 Noboru Yamazoe and Kengo Shimanoe

is not the reducing gas itself but PO2 in the ambient after the reducing gas has been oxidized completely. Resistor-type oxygen sensors working on this principle have been proposed by using oxides such as TiO2,Nb2O5, and MgO-CoO. Among them, one using TiO2 has been successfully incorporated into car engine exhaust control systems in practice. The sensor, fabricated into a well-sintered block of TiO2 with a pair of electrodes inserted, is exposed to car engine exhausts at high temperature (e.g., 1073K). As the resistance decreases or increases stepwise as air/fuel (A/F) ratio crosses the border between lean burn and rich burn, it can be utilized for A/F ratio control. Its share in the market is somewhat small, however, compared with that of its competitor, zirconia oxygen sensors.

1.4 Resistor-type sensors: theoretical aspects For resistive-type gas sensors, a porous assembly of ﬁne particles (mostly grains) of oxide semiconductors should function as a receptor and a transducer. It has long been accepted that grains act as a receptor to gases, while the contacts between the grains act as the transducer which transforms the gas reception into a change in device resistance. However, an understanding of the receptor function and the transducer function involved had remained far from being satisfactory until basic approaches to them began very recently. This section focuses on recent advances in the basic (theoretical) approaches, though the studies are still in progress. 1.4.1 Receptor function and transducer function Oxide semiconductors are known to exhibit unique interactions with some sorts of gases, resulting in the ionosorption of the gases. In the event that the gas in a problem situation has a large electron afﬁnity, such as O2 and NO2, the host semiconductor supplies electrons to the gas to allow it to be 2 adsorbed as anionic species such as O ,O or NO2 . In the event that the gas is low in ionization potential, such as NO, on the other hand, the gas donates electrons to the semiconductor to be adsorbed as cationic spe- þ cies, such as NO . The electrons supplied or given up in these ionosorption processes are transferred from the bulk of the semiconductor to the surface, or vice versa, accompanied by a change in energy band structure (band bending) of the semiconductor. It is well-known that electron transfer from the bulk of n-type semiconductor results in the formation of an electron-depleted layer in the semiconductor. No doubt, an oxide semiconductor sensor, when placed in air, is subjected to the adsorption Fundamentals of semiconductor gas sensors 15

(ionosorption) of oxygen, and its resistance in air (air base) is determined usually from the equilibrium of oxygen adsorption. As very recently revealed with SnO2 sensors, oxygen is adsorbed mainly in the form of O2 in extremely dry air, whereas in the presence of low humidity (0.1% in volume and above), the adsorption in that form is suppressed almost completely by water vapor to be replaced by the adsorption in another form (O ). In practice, it can thus be assumed as a good approximation that the latter form (O ) prevails over the former (O2 ) under usual sensor operating conditions. The sensor is utilized for detecting a target gas coexistent in air by means of a change in the resistance of the device. Target gases fall into two groups: gases which undergo ionosorption (such as NO2) and inflammable gases (such as H2, CO, and C3H8). In cases where ionosorption takes place in addition to that of oxygen, the energy band structure changes accordingly. Usually, however, serious interference often occurs between the ionosorption of the gas and that of oxygen, reducing the resultant change in energy band structure. The key to designing a sensor sensitive to such a gas is discovering how to mitigate such interference. Inflammable gases react with the anionic adsorbates of oxygen. As a result of the reaction, electrons of the adsorbates are returned to the semiconductor, causing the energy band structure to revert to one that corresponds to smaller amounts of oxygen adsorbates. Obviously, response to a gas in this group will be enhanced as the consumption of the oxygen adsorbates is made more efficient. Here, it is of central importance to show how the qualitative understanding mentioned above can be converted into more quantitative ones. For simplicity, let us assume that a sensor device is a porous stack of uniform grains of an n-type oxide semiconductor. It is accepted that each grain plays the role of a receptor, while that of the transducer is played by each contact between grains; that is the most resistive part in the device, so it determines the resistance of the whole device. However, further understanding has been less than straightforward. For some considerable time, efforts were made to understand the receptor and the transducer functions based on the surface space charge layer model and the double Schottky barrier model, as shown by (a) and (b) in Fig. 1.4, respectively. These models, (a) and (b), were guessed at by many researchers as analogies from a metal semiconductor contact diode (see, for instance, Ref. 9). It was assumed that the thickness of the depletion layer (w) should increase as oxygen adsorption as anionic species (typically O ) increases, while it should decrease as the adsorbed oxygen is consumed with an inflammable gas (H2). Correspondingly, the 16 Noboru Yamazoe and Kengo Shimanoe

Figure 1.4 Diagrams of electron depletion for oxide grains and the resistance of contact between grains. (a) Space charge layer model, (b) double Schottky barrier model, (c) regional and volume depletion model, (d) surface conductive grains contact model. double Schottky barrier formed across the contact between grains should change its height, inducing changes in contact resistance and, hence, resistance in the device. Unfortunately, these models were unable to give quantitative information regarding gas response. Shortcomings of the models were made clear recently by our basic approaches, as described below. The receptor model (a) assumes implicitly that the semiconductor grains are sufﬁciently large. In reality, however, they are very tiny (typically about 10 nm in diameter), so the space charge layer can easily extend over the ﬁ entire area of grains; that is, w grows to grain radius (a), at PO2 signi cantly below that in air, PO2 (a). Obviously, a new process of electron depletion has to take place afterward until the grains reach electrostatic equilibrium with oxygen adsorption at PO2(a). A method proposed here is one in which electron depletion is achieved by shifting the Fermi level downward by p kT, as shown in Fig. 1.5.6,7 Here, p is the Fermi level shift as expressed in the unit of kT, where kT is thermal energy. The electrons supplied to the adsorbates in this stage are squeezed out of the grains by increasing p.To distinguish the electron depletion of this type (accompanied by a change in p) from the conventional type one (accompanied by a change in w), these are denoted as volume depletion and regional depletion, respectively. The value of p or w is determined uniquely for given conditions of gas adsorption and semiconductor grains. Importantly, p or w depends on a when the Fundamentals of semiconductor gas sensors 17

(a) (b) PO2 = 0 Ec Ec PO2(I) PO2(I) O–(I) O–(I)

p(II)kT) PO2(II)

– qV(r) O (II) qV(r) PO2(II) p(III)kT) O–(II) p P (III) (III)kT) O–(III) O2 P (III) O–(III) O2 –a 0 a –a/2 0 a/2 r r (c) (d)

Nd Nd P (I) PO2(I) O2 n PO2(II) n

PO2(II)

PO2(III)

P (III) 0 O2 0 –a 0 a –a/2 0 a/2 r r Figure 1.5 Energy band diagrams: (a) and (b) distributions of conduction electrons; (c) and (d) for two kinds of grains different in radius (a or a/2) at steps of increasing . PO2 conditions are otherwise ﬁxed. As shown in Fig. 1.4(c), small oxides are ; usually in a state of regional depletion at low PO2 while those that are usually in a state of volume depletion in base air (the whole area being depleted) and their electronic states are controlled by p. It is noted, however, that more rigorous discussion should be extended in terms of reduced radius (n) rather than of radius (a), as discussed later. The double Schottky barrier model (Fig. 1.4(b)) also turned out to be completely misleading. It focused attention on the electron transport path running through the centers of contacting grains. In reality, however, there are a tremendous number of other transport paths running on the surface of grains, which are free of potential barriers, as shown in Fig. 1.4(d). The electron transport through the contact can thus be achieved by migration or tunneling of the surface electrons, indifferent to the bulk electrons inside. The contact resistance and the device resistance (R) are then inversely proportional to the surface density of electrons, [e]S, as long as the grains are uniform. Device resistance (R) as normalized by that at ﬂat band state 18 Noboru Yamazoe and Kengo Shimanoe

(R0), called “reduced resistance,” is expressed by using the donor density of semiconductor (ND) as follows: R ¼ ND e S (1.1) R0 1.4.2 Response to oxygen (base air resistance) Let us consider a case where oxygen is adsorbed as O on an oxide grain of radius a. The adsorption equilibrium is written as follows: þ ¼ 0 2 ¼ 2 O2 2e 2O KO2PO2 e S O (1.2) Here, KO is the adsorption constant and [O ] the surface concentration of 2 O . Note that [e]S is a variable of the grain. At the same time, we have to consider the electrostatic equilibrium of the grain. Assuming that there is no surface state other than O ,[e ]S and [O ] can be expressed as a function of p, respectively, for volume depletion as follows:15 n o Q n O ¼ SC ¼ N L AðnÞexpð pÞ (1.3) q D D 3 1 ½e ¼ N exp n2 p (1.4) S D 6

Here, QSC is the total surface charge density of the grain, which is assumed to be ascribed solely to [O ] in this case. q is the elementary charge 2 1/2 of proton. LD is the Debye length deﬁned as LD ¼ (εkT/q ND) , where ε is permittivity, and n is reduced radius deﬁned as n ¼ a/LD. A(n) stands for the number of free electrons remaining in the conduction band at p ¼ 0as normalized by NDLD and the surface area of the grain. Assuming Boltz- mann’s distribution law for the tailing of electrons, it is given by the following integral: Z n ð Þ¼ 1 2 1 2 A n 2 R exp R dR n o 6 There are three simultaneous equations, Eqs. (1.2)e(1.4), correlating among three variables, [e ]S,[O ], and p. It is thus possible to determine each variable as a function of KO2PO2. The solution for [e ]S is transformed into normalized resistance through Eq. (1.1). N R S = D ¼ ¼ cðnÞþ ðK P Þ1 2 (1.5) ½ O2 O2 e S R0 a Fundamentals of semiconductor gas sensors 19

S is the shape factor for the semiconductor crystals used; i.e., S ¼ 3 for spheres, 2 for columns, and 1 for plates. Constant c(n) is given by c(n) ¼ (3/n) exp (n2/6) A(n); it increases from unity as n increases; ﬁrst, gradually when n is small and then exponentially afterward. The correlations given by Eq. (1.5) are illustrated in Fig. 1.6, where ð Þ1=2 reduced resistance (R/R0) is related to KO2 PO2 for variously sized grains (LD is assumed to be 3 nm). The linear correlations coincide with the power index (1/2) to PO2, as previously mentioned. Its slope is given by (3/a), indicating that sensitivity to oxygen increases as a decreases. As also indicated in Fig. 1.6, the correlation is bent in the initial region of

PO2 for larger grains where regional depletion takes place. Remarkably, it can be shown that R/R0 is almost independent of a in the regional area. Such correlations have, in fact, been conﬁrmed experimentally. It is also noted that, under a particular condition, oxygen adsorption to form another species (O2 ) also takes place, which is demonstrated by the linear = dependence of R/R on P1 4 in the stage of volume depletion. 0 O2 Notably, the sensor response is related to the kind and amount of oxygen species adsorbed on the surface of the metal oxide semiconductors. The adsorption equilibrium for O and O2 can be discussed as follows, respectively.

O formation: O2 þ 2e ¼ 2O (R1) ð Þ1=2½ ¼ K1PO2 e S O (1.6)

Figure 1.6 Reduced resistance (R/R ) as correlated with ( )1/2/L for devices using 0 KO2PO2 D oxide grains different in reduced size (n). 20 Noboru Yamazoe and Kengo Shimanoe

2 2 O formation: O2 þ 4e ¼ 2O (R3) ð Þ1=2½ 2 ¼ 2 K2PO2 e S O (1.7) Here, K1 and K2 are the oxygen adsorption equilibrium constants of O and O2 , and [O ] and [O2 ] are the concentrations of O and O2 ions, respectively. In this case, the relationship between the oxygen partial pressure and electric resistance is explained using the following Eq. (1.8):16 R ¼ 1 þ 3ð Þ1=2$ 1=2 c K1 PO R0 2 a 2 ( ) = 2 1 2 1 3 1=2 1=2 6ND 1=2 1=2 þ c þ ðK1Þ $P þ ðK2Þ $P (1.8) 4 a O2 a O2

Here, c is a constant. The equilibrium constants K1 and K2 indicate the oxygen adsorption ability on the metal oxide surface as O and O2 . On the base of volume depletion, the relationship between the oxygen adsorption and the electric resistance was investigated. For SnO2, the oxygen adsorption species in dry and wet atmospheres using the relationship between 17,18 the electric resistance and oxygen partial pressure (PO2) was reported. In short, O2 and O adsorb on the surface in dry and wet atmospheres, respectively. In addition, the amount of O was decreased remarkably by adsorption of OH group, resulting that it brings about the deterioration on gas response. Fig. 1.7 shows the relationships between the electric resistance and oxygen partial pressure in dry and wet atmospheres on neat SnO2 (14 nm in diameter) operated at 300 and 350 C. In dry atmosphere, the operating temperature gives different relationship. At 350C, the electric = resistance is directly proportional to P1 4. On the other hand, however, that O2 = = at 300C does not show linearity to both P1 4 and P1 2. This means that O2 O2 2 both adsorption species (O and O ) coexist on surface of SnO2. In wet atmosphere including 0.1 vol% water vapor, the electric resistances at both temperatures decreased as compared with those in dry atmosphere, = but increased in proportion to P1 2. The properties can be understood by O2 moisture effect. Moisture acted as an inhibitor to oxygen adsorption in form of O and O2 , whereas the moisture admitted at elevated temperature acted as a promoter to increase the adsorptive strength of O sites. An increase of the O site population as well as the existence of threshold pressure for oxygen adsorption on the same sites suggests the formation of Fundamentals of semiconductor gas sensors 21

(a) P 1/2/atm 1/2 O2 0 0.2 0.4 0.6 0.8 1

1 300°C 2– –

Ω O and O

0.5 350°C O2– Resistance/10

0 0 0.2 0.4 0.6 0.8 1 P 1/4/atm 1/4 O2 (b) 1.5 Ω 300°C 4 1

350°C 0.5 Resistance/10

0 0 0.2 0.4 0.6 0.8 1 P 1/2/atm 1/2 O2

Figure 1.7 Response of neat SnO2 device to oxygen in dry (a) and wet (b) atmospheres at 300 and 350C.

a sort of surface hydrate, dehydration of which seems to leave O sites behind. The response to oxygen can be understood satisfactorily by using adsorption constant of oxygen, threshold pressure of oxygen, and semiconductor properties of tin oxide in Eq. (1.8). Oxygen adsorption species are different in materials such as receptor loading, surface modiﬁcation, In2O3, and WO3. Table 1.2 shows oxygen adsorption species on each material at 350C. Interestingly, oxygen þ þ adsorption species of Pd-loaded, Sb-doped, and Fe3 or Zr4 -modiﬁed 2 SnO2 are O although the sensors are operated in wet atmosphere. In addition, oxygen on neat In2O3, which is slightly stronger in basicity than 2 SnO2, acts only as O . However, WO3 seems not to have an oxygen adsorption species because surface lattice oxygen is easily formed and active for redox reaction. 22 Noboru Yamazoe and Kengo Shimanoe

Table 1.2 Oxygen adsorption species for each sensor material. Atmosphere Sensor materials (dry or wet air) Temperature (350C)

18) 2 Neat SnO2 Dry O Wet O 19) 2 Pd-loaded SnO2 Dry Wet O 20) 2 Sb-doped SnO2 Dry Wet O 3þ 4þ 21) 2 Fe ,Zr -modiﬁed SnO2 Dry Wet O 22) 2 Neat In2O3 Dry Wet O 23) WO3 Dry Wet Surface lattice oxygen 23) 2 Pd-WO3 (O adsorption, nonreactive)

1.4.3 Response to inﬂammable gases

Simple inﬂammable gases such as H2 and CO react with adsorbed oxygen (O ) in one step, while the supply for the O consumed is the ambient. In a steady state, the following reactions proceed at an equal rate: O2 þ 2e /2O ðR1Þ H2 þ O /H2O þ e (R2) When the rate of the reverse reaction of (R1) is negligible, the surface density of O at the steady state is expressed as follows: ðk P ðaÞÞ O ¼ 1 O2 e 2 (1.9) ð Þ S k2PH2

Here, PO2(a) and PH2 are partial pressures of oxygen in air and hydrogen, respectively, while k1 and k2 are the rate constants of (R1) and (R2), respectively. Eq. (1.9) is a constraint connecting [e ]S and [O ] in this case. Then, the equations for [e ]S,[O ], and p can be solved as previously performed. By using Eq. (1.1), reduced resistance under exposure to H2, Rg/R0 is derived for volume depletion as follows: 1=2 R 3N ðk P ðaÞÞ g ¼ ð Þ= þ ðð Þ= Þ2 þ D 1 O2 c n 2 c n 2 ð Þ (1.10) R0 a k2PH2 This equation is consistent with the power law (1/2) to H2. Reduced resistance in air, Ra/R0, is obtained by substituting PO2(a) for PO2 in Eq. (1.5). The conventional response to H2, Ra/Rg ¼ (Ra/R0)/(Rg/R0), can then be derived, which is expressed as follows when Ra/Rg >>1: 1=2 R Sðk =k Þ = a ¼ 2 1 P1 2ðvolume depletionÞ (1.11) H2 Rg ðaNDÞ

S is the shape factor and equals 3 for spheres; ke1 is the rate constant of ¼ the reverse reaction of (R1), ke1 k1/KO2. The response is thus shown to be Fundamentals of semiconductor gas sensors 23

= linear to P1 2, which accords with the experimental data as shown in H2 2 Fig. 1.9, where (Ra/Rg) is correlated with PH2 instead. The proportionality constant (sensitivity) is promoted with increasing rate constant ratio (k2/ke1) and by decreasing grain radius (a) and donor density (ND). The effects of grain size can thus be rationalized theoretically. However, it should be noted that there are many other inﬂammable gases which react with O in more complex ways. Treatments of the responses to those gases have yet to be undertaken. As mentioned in Section 1.4.2., two types of oxygen adsorption are observed on surface of SnO2. In these cases, the response to inﬂammable gases can be understood as shown in Fig. 1.8. In the case of O (case 1), the electric resistance of SnO2 element is low because of one electron reaction to oxygen atom (xa(O )). The electric resistance decreases by 2 reaction of O adsorbed on SnO2 to H2. In the case of O (case 2), the electric resistance is high because it is the reaction of two electrons, so the change in electric resistance is larger than that in the case 1. The reactions in wet and dry atmospheres correspond to the case 1 and 2, respectively. 1.4.4 Response to oxidizing gases

Let NO2 be an example of an oxidizing gas. It is adsorbed on the grains to form NO2 as follows: þ ¼ ¼½ NO2 e NO2 KNO2PNO2 e S NO2 (1.12)

2– – Xa(O +O )

2– Reaction of O and H2 (case2) 2– O2 + 4e 2O 2– – H2 + O H2O + 2e

– Reaction of O and H2 (case1) – O2 + 2e 2O Resistance – – H2 + O H2O + e

– Xa(O )

Case 2 Case 1 0 0.5 1.0 0 –3 PH ( x 10 ) 2 Figure 1.8 Schematic illustration of the gas response proﬁle as related with O and O2 adsorption species. 24 Noboru Yamazoe and Kengo Shimanoe

2 Figure 1.9 Correlations between gas response (Ra/Rg) and partial pressure of hydrogen ( ) as observed with SnO grains of 12 and 16 nm in diameter at 573K. PH2 2

KNO2 and PNO2 are the equilibrium adsorption constant and partial pressure of NO2, respectively. In base air, oxygen is adsorbed, too, according to Eq. (1.2), so that there are two kinds of adsorbates accommodating electrons transferred from the grain. Through the same procedure used in the previous sections, it can be derived that reduced resistance to NO2 in the stage of volume depletion is expressed as follows: Rg ¼ ð Þþ S ð ð ÞÞ1=2 þ S c n KO2PO2 a KNO2PNO2 (1.13) R0 a a

There is thus linear correlation between resistance and PNO2, with its slope being inversely proportional to the grain size (a). Gas response (Rg/Ra) is derived from Eqs. (1.13) and (1.5), if the grains are already in the stage of volume depletion in air and Rg/Ra >>1, to be as follows: n . o Rg ¼ ð ð ÞÞ1=2 KNO2 KO2PO2 a PNO2 (1.14) Ra The response is independent of a in this case because the dependence of Rg/R0 and Ra/R0 on a is canceled out. If regional depletion prevails in air, however, a totally different situation arises. Now, Ra/R0 is almost independent of a, as stated previously, so that Eq. (1.14) is replaced, approximately, by Eq. (1.15). The response is then inversely proportional to a. Rg ¼ R0 S KNO2PNO2 (1.15) Ra Ra a Fundamentals of semiconductor gas sensors 25

(a) (b) 100 200 300°C

) 150 ) a a 200°C / R

200°C / R g 60 g 400°C 250°C 100

Sensor response (R 50 20 Sensor response (R 300°C

0 0 0 200400 600 800 1000 0 50 100 150 200 250

NO2 concentration / ppb NO2 concentration / ppb

Figure 1.10 Correlations between gas response (Rg/Ra) and partial pressure of nitrogen dioxide as observed with WO3-based devices at various temperatures. (a) Granular WO3 as pyrolyzed from ammonium tungstate, (b) lamellar WO3 with crystallites of about 13 nm in size prepared through a colloidal process.

Devices based on WO3 have been found to be sensitive to NO2,as shown in Fig. 1.9, where granular and lamellar crystals of WO3 are used in Fig. 1.10(a) and (b) respectively. The lamellar crystals with smaller a are seen to be particularly sensitive to NO2, in agreement with Eq. (1.15), being capable of detecting NO2 at 10 ppb. The response is linear to PNO2 at lower operating temperatures. It is suggested that this material allows NO2 to be adsorbed efﬁciently, while keeping O2 adsorption at a minimal level >> (KNO2 KO2), thus imposing the situation rationalized by Eq. (1.15). 1.4.5 Extensions The theory of gas response can be applied or extended to the analyses of other related phenomena of gas sensors, though such work is still in its early stages. For example, the rates of response and recovery have been formulated theoretically.24 It has also been derived that a type of sensitization takes place when semiconductor grains are dispersed with an additive that deprives them of conduction electrons and thus affects the reduction of the effective radius of the grains.25 The theory also provides a useful tool to understand the nature and roles of the metaleoxideesemiconductor contacts involved in semiconductor gas sensors.26 Under conditions where oxide grains are covered with a sufﬁciently large density of adsorbates (surface states), their energy band 26 Noboru Yamazoe and Kengo Shimanoe

Figure 1.11 Energy band diagrams of oxide grain and metal electrode before and after contact under exposure to base air. Note: The band diagrams remain unaltered (pinning) while contact potential is generated in between upon contact. structure is known to remain unaltered, even when they are brought in contact with a metal (pinning). Instead, contact potential (dCP, in volts) is generated across the contact to compensate the work function difference in between, in addition to the conduction band edge difference (dEC) appearing in between, as shown in Fig. 1.11. The expression for volume depletion of dCP is as follows: d ¼ ð4 4 Þ; 4 ¼ 4 þðd ð Þþ Þ q CP q m s q s q s;0 RD n p kT (1.16)

Here, qfm and qfs are the work function values of the metal and semiconductor (f in volts), respectively, and fs,0 is the value of fs at flat band state. The expression dRD(n) kT gives the total lowering of the Fermi level 2 during regional depletion, which is given by dRD(n) ¼ n /(2S), where S is the shape factor. For a resistor-type sensor, dCP acts as a directional barrier to drifting electrons. It reduces the drift mobility of the electrons traveling against it and eventually increases the resistance of the contact involved. It follows that the contacts between electrode metal and oxide grains are more resistive and more gas-sensitive than the other usual contacts between oxide grains. This has been confirmed to be the case with a narrow gap electrodes attached device, in which the aperture between electrodes was as small as 1 mm or below.27 Metalesemiconductor contact also appears to play a key role in the potentiometric gas sensors attached with oxide semiconductors. In these devices, gas response seems to reflect the change of contact Fundamentals of semiconductor gas sensors 27 potential imposed by switching from base air to the target gas ambient, dCP(g) dCP(a), as described later. Through Eq. (1.16) and other relations, it is correlated with the gas response of resistor-type sensors in ideal cases as follows: d ð Þd ð Þ¼4 ð Þ4 ð Þ¼ RT Rg CP g CP a S g s a In (1.17) F Ra Here, R and F are gas constant and Faraday constant, respectively, and RT/F ¼ kT/q. 1.4.6 Nonresistive sensors Various nonresistive gas sensors using semiconductors have been proposed. As described below, these sensors, constructed based on various principles, provide useful information to learn how receptor function and transducer function are generated and combined together into gas sensors, though most of the sensors are yet to be exploited further for use in practice. 1.4.7 Field-effect transistor-type gas sensors 1.4.7.1 Principle The typical structure and characteristic of Pd-gate FET gas sensors are illustrated in Fig. 1.12(a) and (b). As is well-known, a FET, usually attached with a normal metal gate, is a device for controlling drain current by gate voltage applied. Under well-controlled conditions, drain current starts to flow when gate voltage (V) exceeds a threshold voltage (Vth) and, on a further increase 2 in V, it increases proportionally to (V Vth) , as shown in Fig. 1.12(b).Itis endowed with gas sensing ability when the metal gate is attached with an adequate foreign material. If the new gate system modulates the electrical field underneath depending on the gas ambient, the drain current of the device at a fixed gate voltage will change accordingly. Alternatively, actual devices focus attention to Vth and its shift is taken as gas response. The FET gas sensor first proposed was Pd-gate FET; Pd particles were 28 dispersed in the gate region. It responded to H2 and NH3 in air at 423K. Reportedly, the H atoms dissociated from these molecules are dissolved into Pd metal and polarize in the vicinity of the border to the underlying insulator layer (SiO2) to modulate the electrical field underneath. However, with no supporting evidence having been found, this speculation should be reconsidered. Later, various materials were introduced successfully into the gate. Those are typified in three groups: i.e., solid electrolytes, oxide semiconductors, and dielectrics. As observed, combinations of these 28 Noboru Yamazoe and Kengo Shimanoe

(a) VG

ΔV Pd ID SiO 2 V nnDS p-Si

(b)

With H2 Without H2 ΔV

Vth VG

Figure 1.12 (a) Structure of Pd-gate FET. (b) Drain current (ID) characteristics observed: VG, gate voltage, VDS (sourceedrain voltage). materials with the gate metal form gas-sensitive functional systems: half cell, metalesemiconductor contact, or capacitor, respectively.

1.4.7.2 Solid electrolyte-gate ﬁeld-effect transistor Three-phase contact between metal, solid electrolyte, and gas is known to act as an active site for electrochemical reactions (half cell reaction). If the solid electrolyte is a proton conductor, for instance, the following reaction takes place in the presence of H2, and the half cell equilibrium is expressed by the following Nernst equation: þ RT H ¼ 2H þ 2e; F F ¼ In P þ Constant (1.18) 2 M SE 2F H2

The electrical potentials of metal and solid electrolyte are FM and FSE, respectively. The constant is determined by the kinds of materials involved. The same half cell is formed when the proton conductor is placed between the gate metal and the insulator layer of the FET. Thus, FSE is raised by an Fundamentals of semiconductor gas sensors 29

amount as indicated by Eq. (1.18) higher than FM, which is now controlled externally as gate voltage. This means that, at a fixed gate voltage, FSE increases and, hence, the electrical field underneath also increases with increasing PH2. In the alternative mode of operation, Vth shifts down as ’ PH2 increases, following Nernst s equation. Such behavior has been confirmed experimentally with an antimonic acid layer attached device, 29 which responded well to H2 diluted in N2 at room temperature. The response to H2 in air deviated considerably from this behavior because of the occurrence of mixed potential. Similarly, devices sensitive to NO2 or CO2 can be fabricated by attaching þ NaNO2 (Na ionic conductor) or Li2CO3-based composite salt (Liþ ionic conductor) to the gate, respectively.30,31 The response mechanisms involved can be understood in the same way. In the NO2 device, for example, the half cell reaction is expressed as follows: þ RT NO þ e þ Na ¼ NaNO ; F F ¼ InP þ Constant 2 2 M SE F NO2 (1.19)

FSE FM should shift down and so Vth should shift up, with increasing ﬁ PNO2. This behavior has been con rmed experimentally, as shown in Fig. 1.13. The device was fairly sensitive, responding to a few tens ppb NO2 in air, showing a Nernst slope fairly close to that of the one- electron reaction expected. In conventional electrochemistry, a half cell is always combined with another (reference half cell), and its electrochemical equilibrium is investigated through the cell voltage (EMF). In contrast, the half cell of the present device is combined with an FET underneath and its electrochemical equilibrium is investigated through Vth.

1.4.7.3 Oxide semiconductor-gate ﬁeld-effect transistor Oxide semiconductors have been introduced into the gate of FET. A typical example would be the WO3-gate FET, which was sensitive to NO2 in air, as shown in Fig. 1.14.32 Obviously, the high sensitivity originates from the excellent receptor function of WO3 to NO2. In current devices, metalesemiconductor contact is made between the gate metal and ﬁne WO3 crystals, and the resulting contact potential seems to play a decisive role. Owing to the contact potential, FS FM goes up or down with a change in PNO2, according to Eq. (1.16). Here, FS is the electrical potential of WO3. In the same way as the previous devices were treated, the gas 30 Noboru Yamazoe and Kengo Shimanoe

(a) NO2 Gate voltage (VGS)

VG NaNO + WO Source 2 3 Drain electrode electrode Ta 2O5 / SiO2 N-channel N N

P-type

A Source-drain voltage

(b) 400

130°C 350 VDS = 3V ID = 200mA

300 / mV

G 78.9 mV / decade

V (n = 1.0)

250

Air 200 10 100 1000 NO2 concentration / ppb

Figure 1.13 NaNO2-gate ﬁeld-effect transistor (FET) sensor. (a) Construction of NaNO2- gate FET sensor, (b) NO2 sensing characteristics observed.

500

400 92.9 mV /decade (n = 0.9) 150°C 300

/ mV Air G V 119.6 mV /decade 200 (n = 0.8) 180°C Air 100 10 100 1000 NO2 concentration / ppb

Figure 1.14 NO2 sensing characteristics as observed with WO3-gate ﬁeld-effect transistor (FET) sensor. Fundamentals of semiconductor gas sensors 31 response in threshold voltage mode is derived by using Eq. (1.17), given as follows: ð Þ ð Þ¼ RT þ R0 S Vth g Vth a 1n PNO2 1n KNO2 (1.20) F Ra a ﬁ When PNO2 is suf ciently large, Eq. (1.18) is seen to be very similar to

Eq. (1.19), with the response linearly correlated with PNO2 on a semi- logarithmic scale with the same Nernst slope. However, the constants appearing in both the equations have totally different meanings from each other. The constant in Eq. (1.20) mainly reﬂects the sensitivity of the receptor function of the grains to NO2. It determines the position of the semilogarithmically linear correlation along the vertical axis and, so, the lower detection limit of PNO2.

1.4.7.4 Dielectric material-gate field-effect transistor When a layer of dielectric material is introduced beneath the gate metal, a capacitor is formed on the top of the FET, its capacitance varying depending on the dielectric constant and layer thickness of the material. The presence of the capacitor naturally imposes modulation of the electrical field underneath, which is otherwise controlled by the gate voltage only. If the dielectric layer is porous and capable of absorbing a polar molecule gas effectively to change its dielectric constant, the resulting device is made sensitive to the gas through the change in capacitance; Vth moves further away from the air level as the gas partial pressure increases. Based on this principle, the devices sensitive to polar gases (such as water vapor and ethanol gas) have been fabricated fairly successfully by using dielectric materials such as cellulose and its derivatives.33 1.4.8 Oxygen concentration cell type sensors An oxygen concentration cell is constructed by using stabilized zirconia (an O2 ionic conductor) and it is known to work well as an oxygen sensor. If an oxide semiconductor such as SnO2 is deposited between the sensing electrode (Pt) and zirconia (Fig. 1.15(a)), the device is also made sensitive to various reducing and oxidizing gases other than oxygen.34 The response (EMF) to such a nonoxygen gas, starting from 0 in base air, increases or decreases linearly with the increasing logarithm of the partial pressure of the gas (Fig. 1.15(b)), while EMF to a fixed gas ambient varies somewhat drastically with the kind and size of the oxides used. For a considerable time, such a response to nonoxygen gases has been considered to be 32 Noboru Yamazoe and Kengo Shimanoe

Figure 1.15 Oxygen concentration cell type gas sensors attached with oxide semiconductors. (a) Device structure, (b) responses to reducing gas. ascribable to the mixed potential generated at the zirconia/oxide semiconductor interface and, for this reason, such devices have been called “mixed potential” type sensors. The mixed potential is postulated to be generated to H2 in air, for instance, through the following pair of redox reactions: 2 2 O2 þ 4e /2O ; O þ H2/H2O þ 2e However, it is hard to understand why the response is promoted by a decreasing size of oxides (grain size effect) based on this theory. Basic approaches to this group of sensors are highly desired to reveal the fundamental mechanism of gas sensing involved. 1.4.9 Other gas sensors This section describes types of semiconductor gas sensors that have not been mentioned so far.

1.4.9.1 Metaleinsulatoresemiconductor capacitor type sensors The structure of a MIS capacitor and its capacitance versus applied voltage characteristics are shown in Fig. 1.16(a) and (b),35 respectively. A MIS capacitor is obtained if a MIS FET is deprived of the sourceedrain current channel (see Fig. 1.12). To provide the MIS capacitor with gas sensing Fundamentals of semiconductor gas sensors 33

(a) Gold mesh (top electrode) Sensing phase

SiO2

Ta 2O5

Au (backside electrode) (b) 310

300

290 1 ppm NO 280 2 Air 270

260 Capacitance / pF

250

240 200 400 600 800 1000 Applied voltage to p-silicon / mV Figure 1.16 Metaleinsulatoresemiconductor (MIS) capacitor. (a) Structure of MIS capacitor, (b) capacitance versus applied voltage characteristic obtained. ability, foreign materials (solid electrolytes, oxide semiconductors, or dielectrics) are placed under the metal layer, in the same way as in a MIS FET. In these devices, the capacitance depends on the voltage applied to the metal layer (relative to the semiconductor), whereas at a ﬁxed voltage it changes on switching from air base to gas ambient. To keep the capacitance the same, the applied voltage is obliged to shift up or down on changes in the ambient, and this shift is taken as the response of the device to the gas.

1.4.9.2 Diode-type sensors Non-ohmic contact between metal and semiconductor shows a rectifying property, which is utilized in what is known as a “metalesemiconductor contact diode” (Schottky diode). Many researchers have attempted to apply the same principle to gas sensors. Various combinations between metals (Pt, Pd, Ag, etc.) and oxide semiconductors (TiO2, ZnO, etc.) have been chosen to fabricate diode devices. In many cases, the resulting devices 34 Noboru Yamazoe and Kengo Shimanoe

showed a reducing gas-dependent rectifying property; in H2 containing ambient, forward current density was promoted conspicuously with increasing PH2, while reverse current density was also promoted as well, which was unexpected. Such gas-dependent behavior is of sufﬁcient interest from a standpoint of developing gas sensors. At the same time, however, it suggests the need to reconsider the gas sensing mechanism involved. A matter of concern is whether the contacts formed there are, in fact, of the non-ohmic type, as expected. It has been recognized in other semiconductor sensors that the same contact is achieved through generating contact potential instead of undergoing electron transfer, as stated previously. The contact potential can be responsible not only for the rectifying property but also for the promotion of current density, forward as well as reverse, with increasing PH2. Therefore, further careful investigations are needed into this type of gas sensor.

1.5 Future trends Semiconductor gas sensors will become more and more important in the future. Seeds and needs for them, basic approaches needed, and challenges desired are described below as a personal view of the present authors. 1.5.1 Needs and seeds There is a great variety of gases around us of different properties, origin, and concentration. Some are hazardous and should be kept under control, while others may be vital for life or symptomatic of health conditions. Gas sensors are needed for various purposes: safety, amenity, energy saving, health, foods, environmental protection, and so on. As is well-known, the application of gas sensors in practice began with inﬂammable gas alarms to protect people from fatal gas hazards such as gas explosions, incomplete combustion accidents, and exposure to poisonous gases. Fire alarms using a semiconductor gas sensor in combination with a smoke or thermal detector and breath alcohol checkers for preventing drunken driving are also examples of gas sensors used for safety purposes. For the purposes of amenity and energy saving, air quality sensors have been installed in air cleaners, while a pair of sensors sensitive to CO and NO2 has been incorporated into a car autodamper system. Odor sensors and breath odor checkers also belong to this category. Gas sensors are important in other categories, too, though their development is more difﬁcult because the target gases concerned are usually Fundamentals of semiconductor gas sensors 35

Figure 1.17 Microelectromechanical system gas sensor. of very low concentrations. For example, volatile organic compounds are one of the urgent targets; if generated in houses, those may cause sick house syndrome, while some of them are even carcinogenic. Various hazardous gases frequently used in factories, laboratories, and hospitals should be controlled with the use of gas sensors to protect the health of people working there. Sensing of bioactivity-related gases is also important in health and foods. Detection of disease-related gases is drawing increasing attention for medical purposes. Sensory monitoring of air pollutants has been a deep concern to many researchers but, unfortunately, for a variety of reasons this is yet to receive attention. Semiconductor gas sensors, which are endowed with high sensitivity compared with other gas sensors, are, in principle, the best suited for such applications, though a great deal of effort should be put into substantiating new frontiers for gas sensors. As a new seed in gas sensors, microsensors fabricated by using MEMS technology, known as “MEMS sensors,” have recently been exploited extensively, aiming at realizing battery-driven gas sensors. As shown in Fig. 1.17, the gas sensing layer (about 100 100 mm wide and a few tens nm thick) is deposited on a diaphragm, which is suspended over a cavity created within a silicon chip. Electrodes and a heater are printed on the diaphragm beforehand. As a typical feature of such a microdevice, the sensing layer temperature can be changed quickly (within 30 ms), so that the device can be compatible with temperature-programmed operation. This feature would seem to bring about new intelligent functions to gas sensors. Temperature-programmed gas response diagrams, for instance, may be useful for the identification of target gases. 1.5.2 Basic approaches desired Semiconductor gas sensors have so far been developed on the basis of experience and intuition. Tremendous efforts have been devoted to 36 Noboru Yamazoe and Kengo Shimanoe discovering new sensing materials, new ways of materials processing, new types of device, new targets for gas sensing, and so on, putting emphasis on gas sensing performances. This approach, however, is not always so effective for further advances of gas sensors. With the receptor function of small oxide semiconductors having been clarified, there is now a keen need for approaches shedding light on the more basic side of gas sensors. The knowledge thus accumulated will be useful in establishing guidelines for designing semiconductor gas sensors. Matters for further investigation include the following: • establishing methods to characterize and control semiconductive properties, especially the donor density, of oxides; • seeking quantitative correlations between sensitivity data and catalytic oxidation data for a series of inflammable gases; • seeking quantitative correlations between semiconductor properties and gas sensing properties for oxides; • basic analyses of the existing state and the roles of sensitizers; • basic analyses of the effects of mixing one oxide semiconductor with another. Preparation of discrete nanocrystals of oxide semiconductors has become increasingly popular recently. Sensors using nanocrystals of exotic morphology have been fabricated and often shown to exhibit interesting gas sensing properties. Unfortunately, however, origins of such interesting properties have received scant investigation from a basic standpoint, making it difficult to draw on information useful in the design of gas sensors. In fact, nanosized crystallites have already been utilized in practical gas sensors. It would be informative to undertake a critical evaluation of the differences brought about by such a change in morphology. 1.5.3 Challenges There are subjects of research which are worth challenging to progress the innovation of semiconductor gas sensors. Some examples are listed below: 1. Elucidation of control of water vapor effects: Disturbances by water vapor have been a major origin of errors in gas response. Elimination of them upgrades the quality of gas sensing. 2. Verification of ultrasensitive gas sensors: New frontiers of gas sensor applications often demand that they cope with reducing gases at sub-ppm levels. It is necessary, first, to prove that such high-sensitive sensors can be devised. Fundamentals of semiconductor gas sensors 37

3. Contact potential-conscious sensor design: Gas response of a resistor- type sensor seems to be promoted signiﬁcantly by contact potential if a properly designed composite gas sensing layer is used. 4. Exploration to make FET type and oxygen concentration cell type gas sensors more ﬂexible in operating temperature: FET based on silicon cannot function at temperatures higher than c.180C, whereas the cell using zirconia cannot function at temperatures lower than c.550C; neither is able to work in the most important temperature range for gas sensing. Exploration for new semiconductors and new solid electrolytes is desired to eliminate these limitations. References 1. Seiyama T, Kato A, Fujiishi K, Nagatani M. Anal Chem 1962;34:1502. 2. Taguchi N. Published patent application in Japan. 1962. S37-47677, Oct. 3. Xu C, Tamaki J, Miura N, Yamazoe N. Sensor Actuator B Chem 1991;3:147. 4. Rothschild A, Komen Y. J Electroceram 2004;13:697. 5. Rothschild A, Komen Y. J Appl Phys 2004;95:6374. 6. Yamazoe N, Shimanoe K. J Electrochem Soc 2008;155:J85. 7. Yamazoe N, Shimanoe K. J Electrochem Soc 2008;155:J93. 8. Watson J. Sensor Actuator 1984;5:29. 9. Madou M, Morrison SR. Chemical sensing with solid state devices. Boston: Academic Press; 1989. 10. Korotcenkov G. Chemical sensors: fundamentals of sensing materials. New Jersey: Momentum Press; 2011. 11. Yoshioka T, Mizuno N, Iwamoto M. Chem Lett 1991;20:1249. 12. Clifford PK, Tuma DT. Sensor Actuator 1982/1983;3:233. 13. Sakai G, Matsunaga N, Shimanoe K, Yamazoe N. Sensor Actuator B Chem 2001;80:125. 14. Yamazoe N, Shimanoe K. Sensor Actuator B Chem 2011;154:277. 15. Yamazoe N, Shimanoe K. Sensor Actuator B Chem 2011;158:28. 16. Yamazoe N, Suematsu K, Shimanoe K. Sensor Actuator B Chem 2012;163:128. 17. Shimizu Y, Egashira M. MRS Bull 1999;24:18. 18. Yamazoe N, Suematsu K, Shimanoe K. Sensor Actuator B Chem 2013;176:443. 19. Ma N, Suematsu K, Yuasa M, Kida T, Shimanoe K. ACS Appl Mater Interfaces 2015;7: 5863. 20. Suematsu K, Sasaki M, Ma N, Yuasa M, Shimanoe K. ACS Sens 2016;1(7):913. 21. Suematsu K, Uchino H, Mizukami T, Watanabe K, Shimanoe K. J Mater Sci 2019; 54(4):3135. 22. Sun Y, Suematsu K, Watanabe K, Nishibori M, Hu J, Zhang W, Shimanoe K. J Electrochem Soc 2018;167:B275. 23. Hua Z, Yuasa M, Kida T, Yamazoe N, Shimanoe K. Chem Lett 2014;43:1435. 24. Yamazoe N, Shimanoe K. Sensor Actuator B Chem 2010;150:132. 25. Yamazoe N, Shimanoe K. Thin Solid Films 2009;517:6148. 26. Yamazoe N, Shimanoe K, Sawada C. Thin Solid Films 2007;515:8302. 27. Tamaki J, Niimi J, Ogura S, Konishi S. Sensor Actuator B Chem 2006;117:353. 28. Lundstrom€ I, Shivaraman MS, Svensson C, Lundkvist L. Appl Phys Lett 1975;26:55. 29. Miura N, Harada T, Yoshida N, Shimizu Y, Yamazoe N. Sensor Actuator B Chem 1995; 24e5:499. 30. Nakata S, Shimanoe K, Miura N, Yamazoe N. Sensor Actuator B Chem 2001;77:512. 38 Noboru Yamazoe and Kengo Shimanoe

31. Shimanoe K, Goto K, Obata K, Nakata S, Sakai G, Yamazoe N. Sensor Actuator B Chem 2004;102:14. 32. Nakata S, Shimanoe K, Miura N, Yamazoe N. Electrochemistry 2003;71:503. 33. Karube I, Tamiya E, Sode K, Yokoyama K, Kitagawa Y, Suzuki H, Asano Y. Anal Chim Acta 1988;213:69. 34. Lu G, Miura N, Yamazoe N. J Electrochem Soc 1996;143:L154. 35. Zamani C, Shimanoe K, Yamazoe N. Sensor Actuator B Chem 2005;109:216. CHAPTER TWO

Conduction mechanism in semiconducting metal oxide sensing ﬁlms: impact on transduction

N. Barsan,^ M. Huebner, U. Weimar University of Tubingen,€ Tubingen,€ Germany

Contents

2.1 Introduction 39 2.2 General discussion about sensing with semiconducting metal oxide gas sensors 41 2.3 Sensing and transduction for p- and n-type semiconducting metal oxides 47 2.3.1 Modeling of conduction for p- and n-type semiconducting metal oxides in 49 normal conditions (operation in air) 2.3.2 Modeling of the conduction for n-type semiconducting metal oxide: 53 extension to low oxygen concentrations 2.4 Investigation of the conduction mechanism in semiconducting metal oxide 57 sensing layers: studies in working conditions 2.4.1 Sample preparation and experimental conditions 57 2.4.2 Conduction mechanism of p-type CuOdexperimental results 58

2.4.3 Conduction mechanism of n-type SnO2dexperimental results 62 2.5 Conduction mechanism switch for n-type SnO2ebased sensors during operation 66 in application-relevant conditions 2.6 Conclusion and future trends 67 References 67

2.1 Introduction Chemoresistive gas sensors based on semiconducting metal oxides (SMOXs) are very successful, being sold in millions, in applications as diverse as the detection of explosive gas leakages in residential premises, or the control of air intake in car interiors.1 There is a continuous effort to extend their applications in markets as different as indoor air quality or consumer goods (AMS, Austria http://ams.com/eng/Products/Environmental-Sensors/

Gas-Sensors and Sensirion, Switzerland https://www.sensirion.com/en/ environmental-sensors/gas-sensors/multi-pixel-gas-sensors/). After the initial publication of gas sensitive effects on germanium by2; metal oxides e were identified as possible sensitive materials by3 5 and were brought to the market by,6 who founded the largest manufacturer of SMOX sensors: Figaro Engineering (Figaro, Osaka, Japan, http://www.figarosensor.com/). The success of this type of device is based on their good pricee performance ratio; they are • inexpensive (the price range is a few euros per sensor); • easy to use (there is a direct relationship between the concentration of the target gas and the sensor resistance); • very sensitive (generally being able to measure down to a few ppm, or even a few hundred ppb); • very stable (with reported life times extending into decades); • easy to integrate in arrays for more ambitious analytical tasks; and • reasonably low power consumption when realized on micromachined membranes using a pulsed temperature mode (realized by battery operation). The gas detection with SMOX-based gas sensors is, in principle, simple: in air, at temperatures between 150 and 400C, oxygen is adsorbed on the surface of the metal oxides by trapping electrons from the bulk with the overall effect of increasing the resistance of the sensor (for n-type materials) or decreasing it (for p-type materials). The additional occurrences of gases in the atmosphere that react with the preadsorbed oxygen, or directly with the oxide, determine the relative changes of the sensor resistance (sensor signals). From this very naïve picture, one can already get the idea that one has to examine two aspects: the surface reaction taking place between the material and the gases (called the “receptor function”) and the transduction of it into the corresponding changes in the electrical resistance of the sensor. This contribution examines the influence of the conduction mechanism on the transduction of surface reactions into sensor signals. Section 2.2 presents the understanding of the functioning of SMOX-based sensors, and Section 2.3 examines the main differences brought about by the type of conduction of the material. Section 2.4 presents examples of applying simultaneous work function and conductance measurements to the theoretical study of the conduction mechanisms. In Section 2.5, based on experimental results obtained in more realistic conditions (exposure to CO in humid air) and by using the findings from the theoretical modeling we demonstrate that also in practical application a switch of the conduction mechanism is Conduction mechanism in semiconducting metal oxide sensing films: impact on transduction 41 possible. This chapter closes with Section 2.6 which offers a set of conclusions and an outlook for future studies.

2.2 General discussion about sensing with semiconducting metal oxide gas sensors All SMOX-based gas sensors are realized by depositing a sensing layer over an insulating substrate provided with electrodes and a heater. The electrodes are used for the readout of sensor resistance; the heater raises the temperature of the SMOXs sufﬁciently high to allow for their fast and reproducible operation, generally between 150 and 400C. An example is presented in Fig. 2.1. In this example, the sensing layer, in the form of a thick porous ﬁlm, is deposited by applying screen-printing technology onto a planar alumina substrate equipped with interdigitated Pt electrodes on its front, for the readout of the electrical resistance, and a Pt heater on its reverse, which allows the sensor to operate at well-controlled temperatures. All commercial sensors are based on thick porous layers, for reasons that will be given below; in addition to screen printing, other coating technologies (e.g., drop coating)

Sensor device 3.5 mm

layer morphology porous layer with with electrodes large grains 500 μm 1 μm 7 mm 25.4 mm

porous sensing layer

Sensing layer Pt-electrode

Cross section

4.2 mm

Figure 2.1 Design of the sensor substrate used at the University of Tubingen;€ the porous thick ﬁlm sensing layer is deposited on to an alumina substrate, provided with interdigitated Pt electrodes and a Pt heater on the backside allows the operation at well-controlled temperatures. 42 N. B^arsan et al. can be applied. Although the working principles of such devices seem quite simple, the sensing processdwhich includes surface reactions, corresponding charge transfer processes and their translation into variations of the electrical resistance of the sensordis very complex. Fig. 2.2 presents a diagram of the various elements involved in the simple case of CO detection with an n-type SMOX (e.g., an SnO2-based gas sensor).

CO2(gas) CO(gas)

O2(gas) – O– CO– O2 2 2O–

e– e– e–

– – – – O O O– O– O O– O O–

– – – – O O – O – – O O O– O O ∼I R

– – – – – – – – O – O O O O– O O O– O O– O– O O– O–O O–

– – – – – – – O – – O O – – O O – O O – O O O O O– O O– O– O ∼I R

EVac

qV c qVs

Figure 2.2 Sketch representing how the surface reactions are transduced into a measurable signal. Due to the chemisorption of atmospheric oxygen, a depletion layer at the surface of the grains is formed. The presence of reducing gases like CO reduces the negative charge trapped at the surface under formation of CO2. The measurable result is a decrease in the sensor’s resistance (R). The surface reaction and the corresponding conduction situation are indicated by the arrows. Conduction mechanism in semiconducting metal oxide sensing ﬁlms: impact on transduction 43

Fig. 2.2 demonstrates how, due to the chemisorption of atmospheric oxygen, a depletion layer is formed on the surface of the grains comprising the sensing layer. The presence of reducing gases, such as CO, reduces the negative charge trapped on the surface by the formation of CO2. The measurable result is a decrease in the sensor’s resistance (R). The surface reaction and corresponding conduction are indicated by the arrows. The grains of the sensing layer are loosely sintered together; in the example, it is considered that any influence of the surface does not extend into the whole grain, so one can consider that there are two distinctly separate areas: a space charge layer on the surface and, unaffected by exposure to gas, the bulk. In dry air, atmospheric oxygen interacts with the surface of SnO2, acceptor levels are created, and electrons from the conduction band are trapped at these levels, forming molecular and/or atomic oxygen ions. Consequently, the depletion layer appears on the surface of the grains; in the energy band representation, this is formalized as a bending of the upward band, meaning that the electrons need more energy to reach the surface (against the electric field of the negatively charged surface). Hence, the conduction in the sensing layer is controlled by the back-to-back Schottky barriers formed between the grains. It is generally accepted that the CO is reacting with preadsorbed oxygen, forming CO2 that disperses in the atmosphere.7 These surface reactionsdthe ionosorption of oxygen and its consumption by the presence of COdare the chemical basis of sensing; they describe the receptor function of the sensitive material. The charge transfer, associated with the surface chemical reactions, determines the measured effect, namely the resistance change: the reaction of CO with the ionosorbed oxygen decreases the surface negative charge, the consequence being a reduction of the energy barrier height between the grains. That enables a progressively greater number of electrons to flow from one electrode to the other, which translates into a reduction of resistance in the sensor. Assuming that the intrinsic characteristics of the material remain constant, the relationship between the change of the surface charge and the change of the resistance depends on the morphology of the thick film layer. A useful criterion for classification takes into consideration the accessibility of the sensing layer’s bulk to gases, and Fig. 2.3 illustrates a simple distinction between compact and porous layers. In the case of a compact layer, gas interaction only takes place on the geometric surface; the flow of current is only influenced by the thickness of the depletion layer on the surface of the layer. For porous layers, the 44 N. Bârsan et al.

Porous layer Compact layer Gas Product Product Gas

Current flow Current flow

Figure 2.3 Schematic drawing showing the difference between a porous and a compact layer. In case of a compact layer, the gas interaction only takes place at the geometric surface; the current flow is only influenced by the thickness of the depletion layer at the layers surface. For porous layers, the gas can penetrate into the whole layer and by that every single grain is influenced by the surrounding gaseous composition. The current is consequently determined by the barriers between all the grains. gas can penetrate into the entire layer and, in that way, each individual grain is affected by the surrounding gaseous composition. The current is consequently determined by the barriers between all the grains. For compact layers, the bulk is not accessible to gases and the interaction only takes place on the geometric surface (the as-formed electron depleted layer is colored light gray, in contrast to the electron-rich bulk region colored dark gray. Here, the assumption that the constant material properties do not depend on the process by which the layer is formed ensures that, for both type of layer, there are surface and bulk zones). The electrical current therefore flows parallel to the surface and the conduction process takes place in the lower resistive bulk area, with the consequence that it is only indirectly influenced by the modulation of the low resistive cross-section area. This explains why the relative resistance changes for such kinds of layer are low. In the case of porous layers, the gaseous species can penetrate into the bulk, which makes the active surface much deeper. Here, the electrical current is forced to cross the surface by passing from one grain to the next and, accordingly, is directly influenced by the energy barriers between the grains (grain boundary model). These are the main reasons why the best results for n-type metal oxide-based gas sensors are obtained by using porous thick film layers, where the conduction mechanism is controlled by the back-to-back Schottky barriers. Furthermore, the dimensions of the grains in porous layers (d) have to be taken into consideration. There are two casesddepending on the relation between the dimension of the grains, d, and the Debye length, Conduction mechanism in semiconducting metal oxide sensing films: impact on transduction 45

LDdwhich show a different dependency between the conductance and the target gas concentration: C Case 1: grains large enough to have an unaffected bulk area (d >> LD); C Case 2: grains smaller than, or comparable to, the Debye length (d LD). A detailed discussion about the modeling of the two cases is given in Ref. 8 and 9. It must be noted that because of the access of the gases to the whole volume, the interaction can take place in different parts of the sensor device; meaning that, in principle, it is possible to have contributions from the entire sensor and not only from the sensitive material. Fig. 2.4 presents a scanning electron microscopy diagram showing the cross section of an SnO2 porous thick film sensor and the different contact possibilities. In addition to the grainegrain contacts (a), Fig. 2.4 shows additional interfaces that can play a role in sensing and transduction: the graineelectrodeeAl2O3 substrate contact (c) and/or the graineAl2O3 substrate contact (b). The most obvious contribution might be related to (c), due to the fact that the current needs to go through the electrodes and that, due to the noble metal nature of these electrodes, there is a possibility of catalytic effects. The insulating nature of the inert substrate means that (b) is a significantly less probable influencing

(a)

Porous layer with grain boundary model

SnO porous thick film layer m μ 50

Pt-electrode

Grain - AI O -substrate contact ra

Figure 2.4 The different possibilities of gas interaction in the case of porous layers. The gas penetrates into the layer and the interaction can therefore take place at the graine grain boundaries (a), the graineAl2O3esubstrate contact (b), and at the grain electrodeeAl2O3 substrate contact (c). 46 N. Bârsan et al. factor. Furthermore, it was shown that the electrical contribution of the electrodeeSMOX interface is a series resistance that does not change under target gas exposure,9 the influence of which it is possible to minimize by making sure that the number of grainegrain contacts between the electrodes is much greater than 2. This simply implies that the use of gaps between the electrodes is much greater than the grain size. The proven chemical effect of the electrodes10,11 has to be considered as an influencing factor for the average grainegrain sensing unit in the layer. Discussion in the following sections will focus on the part of the conduction process/mechanism in p- and n-type SMOXs. A theoretical discussion (modeling of the conduction) about the different dependencies between the surface chemistry and the corresponding resistance of the layer will be performed. By the use of measurement techniques undertaken during working conditions, the validity of the models will be proven experimentally. In do- ing so, one needs to use a parameter that is directly linked to both surface reactivity and conduction. The ideal candidate is the change in the surface band bending (qDV)of the SMOX because its magnitude under gas exposure is a measure of surface reactivity and also controls the electrical transport from one electrode to the other. Consequently, one needs a technique which is able to directly measure its changes on gas exposure; the well-established Kelvin probe method12 for the measurement of work function changes (DF) was selected for this purpose. Work function changes can be caused by changes in the band bending (qDV), the electron affinity (Dc), or the bulk position of the Fermi level (D(EC,BEF) [ electrochemical potential), as shown in Fig. 2.5 which describes the influence of CO reaction in air on the work function of SnO2. DF ¼ qDV þ Dc þ D EC;B EF (2.1) The latter contribution can be excluded in the temperature range in which SMOX sensors are usually operated because the thermal energy is not sufficiently high for bulk reactions with the atmospheric gases (e.g., oxygen bulk diffusion) to take place. Electron affinity depends on the concentration of surface dipoles, generally linked to surface species related to the reaction with water vapor.13 By ensuring that the electron affinity is constant (Dc ¼ 0), one obtains direct access to the changes in the band bending (qDV) by using the Kelvin probe technique, which can be achieved by keeping the system in very dry conditions. The corresponding changes in the resistance can be easily measured Conduction mechanism in semiconducting metal oxide sensing films: impact on transduction 47

E Vac EVac

χ Φ air χ Φ co

EC,S E C,S qVCO qV E air E D,S Δ ED,S C,B E q V C,B ED,B E E ΔΦ F O- D,B EF E Ea CO in air a

E EV,S V,S

EV,B

x x

Figure 2.5 Energy band representation of SnO2 showing the different contributions to the work function for the example of CO sensing in dry air conditions. Following symbols are used for the different parameters: EVac h vacuum level; EC,S(B) h conduction band at the surface (in the bulk); ED,S(B) h donor levels at the surface (bulk); Ea h surface acceptor levels; EF h Fermi level; EV,S(B) h valence band at the surface (bulk); c h electron afﬁnity; Fair(CO) h work function in air (on CO exposure); qVair(CO) h band bending in air (on CO exposure); qDV ¼ change of band bending; DF h change of work function; and x h distance from the surface. with the sensor device being used. Consequently, one can measure the dependence of the resistance on the surface band bending in different conditions and, therefore, the conduction mechanism can be identiﬁed.

2.3 Sensing and transduction for p- and n-type semiconducting metal oxides In the field of SMOX-based gas sensors, by far the most studied material is the n-type SnO2. Moreover, most of the commercial sensors mar- keted today are based on it, generally in combination with noble metal additives.14,15 The other material used in commercial sensors in applications involving the detection of oxidizing gases is WO3, which is also an n-type semiconductor. This is intriguing because since the early 1980s considerable efforts were directed toward finding alternative materials with a better or different sensing performance, among them p-type oxides, such as Cr2O3 48 N. Bârsan et al. and CuO (see 16e20 for reports on their sensing performance to different gases: H2,O2, EtOH, CO, NO2, etc.). Much lower sensor signals were consistently observedddefined for n-type SMOX sensors as shown in Eq. (2.2) and for p-type SMOX sensors in Eq. (2.3)dwhen compared with SnO2-based sensors in spite of well-known high surface reactivity (R denotes the electrical resistance of the sensor, G denotes the electrical conductance).

Rair Ggas Sn ¼ ¼ (2.2) Rgas Gair

Rgas Gair Sp ¼ ¼ (2.3) Rair Ggas An example for this observation is given in Fig. 2.6, where the EtOH sensing behavior of p-type Cr2O3 (open symbols) is compared with that of undoped SnO2 (ﬁlled symbols) exposed to CO. For both materials, huge changes in the work function (DF, continuous lines with squares) on target gas exposure (EtOH and CO) were measured, indicating a high surface reactivity (change of band bending). In the case of n-type SnO2, these changes are translated into rather “large” sensor signals (dotted line with ﬁlled circles). For p-type Cr2O3, similar changes resulted in much lower levels of signal (dotted line with open circles). The reason for this

CO conc. [ppm] 0 20 40 60 80 100

0,00 14

12 –0,05 10 ΔΦ (Cr O ) –0,10 2 3 S (Cr2O3) 8

[eV] ΔΦ –0,15 (SnO2) 6 S ΔΦ (SnO2) 4

–0,20 Sensor signal

2 –0,25 0 01020304050 EtoH conc. [ppm]

Figure 2.6 Comparison between p-type Cr2O3 and n-type SnO2. For similar changes in the work function on exposure to EtOH and CO, respectively, the n-type material shows much higher changes in the resistance (sensor signal S). Conduction mechanism in semiconducting metal oxide sensing ﬁlms: impact on transduction 49 was recently unveiled when it was demonstrated that large changes in the surface band bending (qDV) do not result in large changes in resistance (sensor signals) because of the conduction mechanism.21,22 To understand and explain this huge difference in the relationship between the surface chemistry and the changes in conductance, one must look in more detail at the conduction processes in these materials. The focus will be, ﬁrst, on the modeling of the conduction for p-type metal oxides to determine the relationship between resistance and band bending in normal operational conditions. Subsequently, n-type materials will be examined for conditions in which the conduction mechanism changes.

2.3.1 Modeling of conduction for p- and n-type semiconducting metal oxides in normal conditions (operation in air) 21 Even if the initial modeling were based on results obtained on Cr2O3, the large grain size of that material would have the effect of making the weight of surface phenomena even less signiﬁcant because of the possible conduction contribution of the bulk. Accordingly, to simplify the case under investigation, CuO was used as a prototype p-type metal oxide for the gas sensing performance in response to CO; its grain size making it feasible to consider that the surface plays the dominant role in conduction.23 The validity of the ﬁndings, although, is not limited to CuO. The p-type semiconducting behavior of CuO is related to the presence of acceptor levelsdattributed to copper vacanciesdin the band gap, which determine the appearance of holes in the valence band. The adsorption of oxygen on the surface of CuO is considered to be at the origin of CO sensing and can be described by Eq. (2.4):

1 þ Oair þ S 4O þ h (2.4) 2 2 A ðadÞ air where O2 represents atmospheric oxygen, SA an adsorption site for oxygen, þ OðadÞ the resulting chemisorbed oxygen species, and h the created hole in the valence band. The interaction of atmospheric O2 with the surface of the SMOX determines the formation of acceptor levels, and the electron transfer from the valence band to the surface leads to the formation of ionosorbed oxygen species resulting in upward band bending. The negatively charged surface is compensated by an increased hole concentration in the valence band that determines the formation of an accumulation layer. This is a very important difference when one compares the case of p-type 50 N. B^arsan et al. materials with n-type materials: in p-type, the conductivity of the surface increases because of the adsorption of atmospheric oxygen. This situation, represented by energy bands, is shown on the left-hand side of Fig. 2.7. The effect of CO exposure, very similar to the general reaction mechanism for SnO2 explained in Section 2.2, is the consumption of ionosorbed oxygen species that determines the reduction of negative charge trapped at the surface: gas þ e þ þ/ gas þ CO OðadÞ h CO2 SA (2.5) gas e where CO represents the carbon monoxide in the gas phase, Oð Þ pre- þ gas ad adsorbed oxygen species, h a hole in the valence band, CO2 the formed product, and SA a free adsorption site for oxygen. As a consequence, the hole concentration near the surface decreases; in energy terms, this situation is described by a decrease in the surface band bending. Hence, the conductivity at the surface decreases and the overall sensor resistance increases. Fig. 2.8 illustrates the differences in the conduction mechanisms between “similar” porous layers of p-type and n-type materials. The “similarity”d meaning comparable grain size, morphology, and parameters of the depletion/accumulation layersdis considered to focus on the conduction mechanism only. The left-hand side of Fig. 2.8 describes an n-type SMOX for a porous layer consisting of loosely sintered grains with a radius larger than the Debye length (not fully depleted). In this case, one can

EVac EVac χχ Φ Φ air CO EC.S EC.S qVco qVair Δ E q V C.B E C.B

E ΔΦ F E E V.S F EV.S CO EV.B E E Ea V.B a

X X Figure 2.7 Energy band representation for a p-type semiconducting metal oxide material. The reaction of CO as target gas with the p-type material causes a decrease in the band bending (qDV) and therefore changes in the work function (DF). The used symbols for the different parameters are equal to Fig. 2.5. Conduction mechanism in semiconducting metal oxide sensing ﬁlms: impact on transduction 51

n-typeSMOX p-typeSMOX

O O O O O O O O O O O O O O O O O OO O O O O O O O O O O O O O

O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O

Depletion layer Accumulation layer

Figure 2.8 Cartoon-like illustration of the conduction processes and the corresponding energy band representation for an n-type semiconducting metal oxide (SMOX) material (left) described by a depletion layer and for a p-type SMOX material (right) with an accumulation layer. differentiate between a gas sensitive surface depletion layer (upward band bending, light gray) with a large electrical resistance and an unaffected bulk region (dark gray) with a lower electrical resistance. The electric current through the layer from one electrode to the other is therefore determined by the concentration of electrons (nS) having sufficient energy to overcome the potential barrier (qVS) between the grains (back-to-back Schottky barriers). The dependence between this concentration and the surface band bending can be described by a Boltzmann distribution by assuming that the Schottky approximation is valid: qV n ¼ n exp S (2.6) s b kT where nb represents the electron density in the bulk and kT the thermal energy (z0.05 eV). For the conductance, one can consequently write qV G fexp S (2.7) n kT As shown in Fig. 2.7, the upward band bending in the case of p-type MOXs determines the formation of an accumulation layer for holes. Accordingly, the conductivity in the surface space charge layer increases in comparison with the bulk, and conduction takes place differently compared with that described by the depletion layer. The current will now flow through the accumulation parallel to the surface and also through 52 N. Bârsan et al. the bulk; this situation can be described by two resistors in parallel. The latter contribution from the bulk depends on the nature of the material and the morphology of the layer. It is obvious that larger grains have a higher bulk influence, which makes the surface effectdand, therefore, the sensingdless important. In this case (e.g., for Cr2O3 with rather large grains), a complex relationship between the conductance/resistance and the band bending was obtained.21 A much simpler relationship is devised by ensuring that the grains are quite small (e.g., as in the case of CuO (z25 nm)),23 so that the contribution to the conductance of the bulk can be ignored. Hence, one can assume that the conduction process is now dominated by the average hole concentration in the accumulation layer fe (Gp pS). Considering that also in this situation the Boltzmann statistics are valid, e the average hole concentration pS can be easily calculated by using a one- dimensional approach.

Zx0 ð Þ e ¼ 1 $ qV x ps pb exp dx (2.8) x0 kT 0 where pb represents the hole density in the bulk and x0 the width of the space charge layer. To evaluate the integral in Eq. (2.8), one must solve the Poisson equation for the accumulation layer if the conductance is determined by the holes. After the first integration of Poisson’s equation, one obtains24 rffiffiffiffiffiffiffiffiffiffiffiffi dV ðxÞ 2kTp qV ðxÞ ¼ b$exp (2.9) dx εε0 2kT where V represents the potential at a certain point x and εε0 the relative permittivity of the material. By using the definition of the Debye length pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 LD ¼ kTεε0=q pb and the boundary conditions that V ¼ VS for x ¼ 0, the following relationship between the distance from the surface, x, and the potential V(x) is obtained for the second integration: pffiffiffi qV qV ðxÞ x ¼ 2$L $ exp S exp (2.10) D 2kT 2kT Now, the integral in Eq. (2.8) can be calculated by changing the variables and by using Eq. (2.10). For the average concentration of holes in the surface space charge layer, one obtains Conduction mechanism in semiconducting metal oxide sensing films: impact on transduction 53

qV ep ¼ p exp S (2.11) s b 2kT Hence, one can write for the dependence of the conductance and the surface band bending: qV G fexp S (2.12) p 2kT Comparing the latter expression with the corresponding one for n-type porous thick film layers (Eq. 2.7), one can clearly observe that the same surface chemistry (same change in the surface band bending) is translated differently into a change of the conductance/resistance depending on what kind of material is used. Considering the sensor signal as the relative change of the sensor resistance due to exposure to the target gas, Sn,p, one obtains that the signal of a p-type is simply the square root of the signal for the n-type MOX: pffiffiffiffiffi Sp ¼ Sn (2.13) This fact clearly shows why the use of p-type materials as chemoresistive gas sensors is not optimal. Although these materials may be highly reactive to the target gases, the output electrical signal is quite low, as already shown in Fig. 2.6 for Cr2O3. To calculate the band bending changes from the changes in the resistance, one has to use the following dependency: R qDV ¼2kT$ln gas (2.14) Rair More details about the calculations can be found in Ref. 23. The classical, state-of-the-art preparation technology for SMOX-based gas sensorsdthick, porous sensing layersdis not the best choice for p-type materials. In their case, the direct readout of the changes in the surface band bending would be more efficient; in the case of a resistive readout, thin, compact films with electrodes deposited on the top would be more appropriate.

2.3.2 Modeling of the conduction for n-type semiconducting metal oxide: extension to low oxygen concentrations

As already mentioned, for the case of n-type SnO2 porous layers, the conduction mechanism in an oxygen containing background is determined by the appearance of a depletion layer at the surface of the grains. The negative surface charge related to ionosorbed oxygen species is compensated by a 54 N. Bârsan et al. positive space charge layer near the surface, resulting in an upward band bending. In cases where the dimension of the grains is larger than the Debye length, one can distinguish between a rather resistive space charge layer on the surface and a bulk area with a lower resistance. The conduction is therefore controlled by the barrier height on the surface of the grains. Only the electrons which have sufficient energy to overcome the back-to-back Schottky barriers between the grains can move from one electrode to another. The diagram of the conduction process in the depletion layer in Fig. 2.9 helps achieve a better understanding. The dependence of the conductance and the surface band bending can be easily expressed by Eq. (2.7), assuming that the Schottky approximation is valid and that the bulk donors are fully ionized. By decreasing the amount of oxygen in the background (or by increasing the concentration of the reducing gases), one consequently also decreases the initial upward band bending on the surface of the grains; in certain conditions, it is possible to reach a flat band situation. That means that there are no energy differences between the surface and the bulk, as well as the fact that the concentration of the free charge carriers is constant. A similar situation is possible also in air when the Debye length exceeds the grain size: fully depleted grains. There, the position of the Fermi level relative to the minimum of the conduction band does not correspond to the Fermi level of the bulk conditions.9 If the oxygen concentration in the atmosphere were lowered, the flat band situation could most probably be reached in the absence of oxygen, if one considers that the full height of the upward band bending is only

Depletion layer n-type SMOX Accumulation layer n-type SMOX – – – O O – O– O – O– O– O O– O– O–O O–

– O – – – – – – – O O – O O O O O O– O O– O– EE

EF –qVs Ec.s Ec.s qVs

Figure 2.9 Cartoon-like presentation of the conduction processes for n-type SnO2 in case of depletion layerecontrolled and accumulation layerecontrolled model. SMOX, semiconducting metal oxide. Conduction mechanism in semiconducting metal oxide sensing ﬁlms: impact on transduction 55 determined by the ionosorption of oxygen, it means that there are no intrinsic electron traps on the surface. If the band bending is further decreased (e.g., by exposure to reducing gases that will form surface donors25), one would record a downward band bending at the surface, meaning the formation of an accumulation layer and, therefore, the conduction mechanism will change. This situation is presented in the right-hand side of Fig. 2.9. The easiest way for the electrical current to travel from one electrode to another is now through the accumulation layer on the surface (lowest resistivity). Hence, one can also assume, as in the case of p-type MOXs, that the conductance is simply proportional to the average electron concentration in the accumulation layer (Gn f ~nS). The challenge is now to ﬁnd a relationship which describes the dependence between the changes in the band bending and the changes in the resistance for the accumulation layer. The procedure is very similar to that shown previously for the p-type material and is described in detail in Ref. 26. By assuming that also in the accumulation layer conditions the Boltzmann statistics are valid and that one can use a one-dimensional approach, one can write for the average electron concentration:

Zx0 1 qV ðxÞ enS ¼ $ nb exp dx (2.15) x0 kT 0 The only difference compared with the accumulation layer for the p-type SMOX is that the electrons are the charge carriers and that the accumulation is therefore described by a downward band bending. Again, here one has to solve the Poisson equation24; with its solution, the average electron concentration ~nS can be calculated. One obtains qV en ¼ n $exp S (2.16) S b 2kT The dependence of the conductance on the surface band bending in the case of the accumulation layer can therefore be described by qV G fexp S (2.17) n 2kT 56 N. B^arsan et al.

By comparing Eqs. (2.17) and (2.7), it becomes obvious that the impact of the surface band bending on the conductance is very different depending on the condition (depletion vs. accumulation layer). If the conduction is described by the accumulation layer, the relative change of the resistance is simply the square root of the relative change in relation to the depletion layer model, assuming that in both cases one measures the same surface band bending changes: pffiffiffiffiffiffiffiffi Sacc ¼ Sdep (2.18) The latter expression implies that the “largest” signals are obtained in the case of the depletion layer model. The effect of the surface chemistry on the resistance changes becomes weaker where the conduction moves into the accumulation layer. The analytical solution (Eq. 2.16) obtained for the average electron concentration in the accumulation layer is only valid in conditions where one can use the Boltzmann statistics. If the conduction band edge at the surface crosses the Fermi level, this assumption is no longer valid. One then has to use the FermieDirac statistics instead of the Boltzmann statistics to determine the dependence of the electron concentration on the surface band bending. The average electron concentration in the accumulation in this case can be calculated numerically.26 By comparing the trend of the analytical and the numerical solutions, both describing how the average electron concentration in the accumulation layer depends on the surface band bending, the following statements can be made: • both trends are similar (same slope) up to around 0.3 eV after the crossing of the surface conduction band edge with the Fermi level position; • with further increasing of the surface band bending, the trends are strongly divergence, which reflects the lack of appropriateness of the Boltzmann approximation. The slope of the “correct” numerical solution is getting lower (a smaller coefficient than (2kT) 1 in the exponent in Eq. [2.17]), indicating that the influence of the band bending on the resistance is becoming weaker. The latter calculations in Eqs. (2.15) to (2.17) show that the conduction mechanism for an n-type SnO2 sensor might change from one controlled by a depletion layer to one dominated by transport through the accumulation layer, depending on the operational conditions. This fact has to be borne in mind, especially if the concentration range to be explored is very large. Conduction mechanism in semiconducting metal oxide sensing films: impact on transduction 57

2.4 Investigation of the conduction mechanism in semiconducting metal oxide sensing layers: studies in working conditions

The ideas and models presented in Section 2.3 are applied to SnO2 and CuO as model systems for n- and p-type SMOXs. 2.4.1 Sample preparation and experimental conditions

The n- type semiconducting SnO2 powder was synthesized by a conventional wet chemistry solegel procedure (SnCl4(aq) and NH3(aq)) followed by a calcination treatment at 1000C for 8 h.27 For the highly crystalline p-type CuO nanoparticles, a soft chemistry route was employed using a mixture of copper acetate, oleic acid, and trioctylamine.23 The investigated porous thick film layers were obtained by using the automatic screen- printing procedure. Therefore, the powders were mixed with an appropriate amount of an organic vehicle to obtain a homogenous paste which was subsequently printed onto the alumina substrates. For the electrical readout, the substrates are provided with interdigitated Pt electrodes, and a Pt heater on the backside allowed the operation at well-controlled temperatures (see Fig. 2.1). To remove the residual organic solvent, the sensors were finally heated in a moving belt oven (SnO2: 400e600 C; CuO: 300e450 C). To investigate the conduction mechanism in the sensing layers, simultaneous DC resistance and work function change measurements were taken in working conditions using the Kelvin probe technique (McAllister KP 6500K Probe). The latter is a noncontact, nondestructive method which measures the changes in the contact potential differences (CPDs) between the sensor and a vibrating reference electrode. Variations in the CPD induced by the changes in the surrounding atmosphere (e.g., CO exposure) represent the changes in the material’s work function.12 DCPD ¼ eDF=q (2.19) By choosing conditions in such a way that possible contributions from the electron affinity (Dc) to the work function changes can be ignored (very dry conditions excluding influences of surface dipoles from humidity), the changes in the surface band bending can be directly measured as changes in the contact potential difference (DCPD). In these circumstances, one can directly correlate the changes in the surface band bending with the corresponding sensor resistance change. It is important to note that both measured parameters are average values corresponding to an average sensing unit of the 58 N. Bârsan et al. layer, which includes all influences from grain size dispersion, influence of substrate and electrode, etc.

2.4.2 Conduction mechanism of p-type CuOdexperimental results Simultaneous DC resistance ([_Keithley_2000] multimeter) and work function change measurements on exposure to CO (10, 30, 50, 70, and 100 ppm) of a CuO-based porous thick ﬁlm gas sensor were performed at 150C in dry air conditions. The time dependencies of the resistance and the CPD are presented in Fig. 2.10. One observes that the resistance increases on CO exposure, whereas the work function decreases. This indicates that there are fewer holes in the accumulation layer and a decrease of the upward band bending occurs (see also Fig. 2.7). One expects that, in dry conditions, CO reacts with preadsorbed oxygen, resulting in the cancellation of a hole and the formation of CO2 as described by Eq. (2.5). In these circumstances, the measured changes in the work function on exposure to CO should only be caused by changes in the band bending. To prove this assumption, the different contributions which may cause changes in the work function on increasing CO concentrations are shown in Fig. 2.11. The changes of the work function (DF, dark gray with stars) are directly measured, and the changes in the band bending (qDV, black line with dots) are extracted from the resistance changes by using the relationship for the dependence of the resistance and the surface band

Resistance CPD

0.3 )

0.2 Ω 100k 100ppm CO 70ppm CO 50ppm CO 30ppm CO 10ppm CO

0.1 CPD (V) Resistance (

0.0

10k 0246610 Time (h) Figure 2.10 Simultaneous contact potential differences (CPDs) and electrical resistance changes of a CuO sensordoperated at 150Cddue to exposure to different concentrations of CO (10, 30, 50, 70, and 100 ppm) in dry air conditions. Conduction mechanism in semiconducting metal oxide sensing ﬁlms: impact on transduction 59

qΔV 0.02 ∆Φ Δχ

0.00

–0.02 ∆E (eV)

–0.04

–0.06 0 100030405060708090100110 CO conc. (ppm) Figure 2.11 Changes of band bending (qDV), work function (DF), and possible changes of the electron affinity (Dc) with increasing CO concentrations (10, 30, 50, 70, and 100 ppm) of the CuO sensor in dry air condition operated at 150C. DE represents the changes of all contributions in the unit of eV. bending (Eq. 2.14). The changes of the electron affinity (Dc, open squares) are calculated according to the following equation: Dc ¼ DF qDV (2.20) One clearly observes that no changes occur in the electron affinity during the reaction of CO with CuO in dry air conditions; all changes in the work function are caused by changes in the band bending. The fitting curve describing the dependence between the sensor signal (Sp ¼ Rgas/Rair) and the corresponding changes in the work function/ band bending is given in Fig. 2.12. The value for the slope, as obtained, fairly accurately reflects the dependence gained from the theory (experimental value of 2.000 0.034 and theoretical value of 2). The experimental results show a good match to the theory. In addition, it was becoming clear that the use of p-type materials as chemoresistive gas sensing materials is not optimal. The same surface chemistry (same band bending) results in a much lower sensor signal for the p-type material compared with n-type materials, where the conduction is described by a depletion layer model (upward band bending). Furthermore, the reaction of CO with preadsorbed oxygen species is supported because no changes in the electron affinity are observed. To extend the findings to more realistic orientated conditions, similar experiments in a background of 50% relative humidity (25C) were 60 N. Bârsan et al.

Sensor signal 1 ∆Φ Sp = exp 2.000 +– 0.034KT

0.0 0.1 0.2 0.3 0.4 0.5 0.6 ∆Φ/2kT

Figure 2.12 Fitting curve describing the dependence between the sensor signal (Sp) and the corresponding changes in the work function. The as-obtained value for the slope reﬂects quite well the dependence gained from the theory.

0.00

–0.02

–0.04 ∆E (eV)

–0.06 qΔV ∆Φ –0.08 Δχ

0 20 40 60 80 100 CO conc. (ppm) Figure 2.13 Changes of band bending (qDV), work function (DF), and possible changes of the electron affinity (Dc) with increasing CO concentrations (10, 30, 50, 70, and 100 ppm) of the CuO sensor in 50% relative humidity (25C) operated at 150C. performed. Fig. 2.13 presents the different contributions: the measured changes in the work function (stars), the calculated band bending changes (black with circles) from the measured resistance changes, and the extracted electron affinity (squares) changes on CO exposure in the humid background. There, the situation is completely different compared with the Conduction mechanism in semiconducting metal oxide sensing films: impact on transduction 61 experiments in dry air: significantly larger changes in the work function are observed, whereas, at the same time, the decrease in the band bending is much smaller. This implies that, on CO exposure in the presence of humidity, a strong decrease occurs in electron affinity. For a better understanding of the difference observed between Figs. 2.11 and 2.13, the influence of the humidity itself and its effect on the different contributions is depicted in Fig. 2.14. The exposure to water vapor not only determines a large decrease in the band bending (increasing resistance, circles) but also an increase in the electron affinity (squares), which indicates a change in the concentration of surface dipoles. The reaction of water with the surface of CuO can be expressed as follows (see also25): e þ gas þ þ þ4 þ e þ OðadÞ H2O 2CuCu h 2 CuCu OH SA (2.21) gas A water molecule from the atmosphere (H2O ) reacts with preadsorbed e oxygen ions (OðadÞ) and two Cu sites (2CuCu) on the surface under the for- þ e mation of two terminal hydroxyl groups 2 CuCu OH . The appearance of the two terminal hydroxyl groups is responsible for the increase in the electron affinity (formation of local surface dipoles) and the cancellation þ of a hole (h ) determines the decrease in the band bending. SA is the freed adsorption site for chemisorbed oxygen. Consequently, there is competition between CO and H2O for oxygen ions as reaction partners in the presence of humidity. This explains the

0.06

Δ 0.03 q V ∆Φ 0.00 Δχ

–0.03

∆E (eV) –0.06

–0.09

–0.12

–0.15 020406080 Relative humidity (%) Figure 2.14 Changes of band bending (qDV), work function (DF), and possible changes of the electron affinity (Dc) of the CuO sensor on exposure to humidity levels (10%, 30%, 50%, and 70% relative humidity @25C) at an operation temperature of 150C. 62 N. Bârsan et al. observed behavior in Fig. 2.13. The effect of CO exposure in humid conditions is reduced (smaller sensor signals); and the buildup of the dipoles is hindered due to fewer adsorption sites for water vapor, which determines the monitored decrease in the electron affinity. This example demonstrates how one can identify the sensing mechanism of CO and CuO in the presence of humidity by using working condition DC resistance and work function change measurements in combination with appropriate modeling of the conduction.

2.4.3 Conduction mechanism of n-type SnO2dexperimental results

The experiments were performed on an SnO2-based gas sensor operated at 300 C. Investigations were made into the inﬂuences of CO and H2 in different oxygen backgrounds, and that of oxygen itself on the resistance and the band bending. In Fig. 2.15, the measured changes of the resistance and the CPD during stepwise increasing oxygen concentrations from 0 (N2 atmosphere) up to 2500 ppm are shown. One observes a steep increase in both resistance and work function at the lower concentrations due to the adsorption of oxygen, resulting in ionosorbed oxygen species on the surface; a form of saturation occurs as oxygen reaches 2000 ppm.

Resistance CPD 1M

0.00 2 2 2 ) 2 –0.03 Ω 2

–0.06 2000ppm O 2500ppm O 1000ppm O 500ppm O 2 CPD (V) 300ppm O –0.09 Resistance ( 100K 100ppm O –0.12

0 5 10 15 20 25 Time (h) Figure 2.15 Simultaneous contact potential differences (CPDs) and electrical resistance changes of an SnO2 sensordoperated at 300 Cdwith stepwise increasing oxygen amount (100, 300, 500, 1000, 2000, and 2500 ppm). Conduction mechanism in semiconducting metal oxide sensing ﬁlms: impact on transduction 63

(a) (b) Resistance CPD 0.8 10M 0.8 10M

0.7 0.7 1M 1M 0.6 0.6 ) ) 0.5 10ppm CO 0.5 Ω Ω 30ppm CO

100k 70ppm CO 100ppm CO 100k 0.4 0.4 10ppm CO

0.3 0.3 10k 10k CPD (V) CPD (V) 0.2 0.2 Resistance ( Resistance ( 30ppm CO

70ppm CO 0.1 0.1 100ppm CO 1k 1k

0.0 0.0

–0.1 100 –0.1 100 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 Time (h) Time (h) Figure 2.16 Simultaneous contact potential differences (CPDs) and electrical resistance changes of an SnO2 sensordoperated at 300 Cdduring exposure to four pulses of CO (10, 30, 70, and 100 ppm) in the absence of oxygen (a) and in a background of 22,000 ppm of oxygen (b).

Fig. 2.16 illustrates the behavior of the resistance and the CPD on exposure to four CO pulses (10, 30, 70, and 100 ppm) in the absence (Fig. 2.16(a)) and in the presence (Fig. 2.16(b)) of 22,000 ppm of oxygen. A huge drop is observed in both the resistance and the work function due to exposure to CO; except for the first pulse (10 ppm), the equilibrium state in the work function is reached in the allotted 3 h of CO exposure. The recovery process, however, is very slow for both parameters; the baseline could not be reached again in the 3 h allowed for recovery. The presence of oxygen in the background determines a higher baseline resistance (formation of ionosorbed oxygen) and a decrease in the relative changes of both the resistance and the work function in comparison with the absence of oxygen. The response and recovery times in the presence of oxygen are considerably more rapid. Fig. 2.17 illustrates the time dependence of the resistance and the CPD of a similar experiment using five pulses of H2 (10, 20, 30, 50, and 100 ppm) instead of CO in the absence of oxygen (Fig. 2.17(a)) and in a background of 22,000 ppm of oxygen (Fig. 2.17(b)). In the case of hydrogen, the drop in the resistance and in the work function in the absence of oxygen is much more dramatic (the material becomes almost conductive). The response and recovery times are more rapid compared with CO. In the presence of oxygen, as expected, increases in the baseline resistance and the decrease of the signals were observed. Fig. 2.18 presents an overview of all the results obtained by simultaneous DC resistance and work function change measurements, including similar 64 N. Bârsan et al.

(a) (b) Resistance CPD 1.4 10M 1.4 10M

1.2 1.2 1M 1M 1.0 1.0 ) ) 100k 10ppm H 100k Ω Ω 20ppm H

0.8 0.8 30ppm H 50ppm H 100ppm H

0.6 10k 0.6 10k

CPD (V) 0.4 CPD (V) 0.4 1k 1k 10ppm H 20ppm H Resistance ( 0.2 Resistance ( 0.2 30ppm H 50ppm H 100ppm H 100 100 0.0 0.0

–0.2 10 –0.2 10 0 5 10 15 20 25 30 0 5 10 15 20 25 30 Time (h) Time (h) Figure 2.17 Simultaneous contact potential differences (CPDs) and electrical resistance changes of an SnO2 sensordoperated at 300 Cdduring exposure to ﬁve pulses of H2 (10, 20, 30, 50, and 100 ppm) in the absence of oxygen (a) and in a background of 22,000 ppm of oxygen (b).

10M O2 H (0ppm O ) qV 2 2 1M R ∼ S exp CO (0ppm O ) 1.0kT 2 H2 (200ppm O2) ) 100k Ω CO (200ppm O2) qVS R ∼ exp H (22000ppm O ) 2.35kT 2 2 10k CO (22000ppm O2)

Resistance ( qV R ∼ exp S 3.80kT 100 Flat band situation

10 0.2 0.0 –0.2 –0.4 –0.6 –0.8 –1.0 –1.2 –1.4 q∆V (eV) Figure 2.18 Dependence of the resistance and the corresponding band bending changes (qDV). The flat band situation is denoted as the situation in a dry N2 background. The three different regions/models are shown with the lines. experiments in a background of 200 ppm of oxygen. There, the resistance and the corresponding band bending changesdextracted from the changes in the work functiondare plotted semilogarithmically. As a reference, the situation in nitrogen was chosen (qDV ¼ 0). The existence of three different areas with a seamless transfer in between each other is obvious. Each of them can be accurately fitted by a proper exponential dependency of the resistance and the corresponding band bending. The calculated slopes are ((1.0 0.1) kT) 1, ((2.35 0.12)kT) 1, and ((3.80 0.05)kT) 1, respectively. Conduction mechanism in semiconducting metal oxide sensing films: impact on transduction 65

The trend observed in the experimental data is in line with the theoretical modeling presented in Section 2.3.2. The conduction moves from a mechanism controlled by the presence of the depletion layer (theoretical value: 1; experimental value: 1.0 0.1) to one controlled by transport through the accumulation layer where the Boltzmann statistics are still valid (theoretical value: 2; experimental value: 2.35 0.12) to the extreme case in which the Fermi level extends deep into the conduction band on the surface. It could be demonstrated in the theory26 that, in this area, the value should increase above 2, reflecting that the influence of the surface band bending on the resistance decreases. This decrease is also supported by the experimental/phenomenological parameter showing a value of 3.80 0.05. The latter results correlate the measured resistances with the measured changes in the work function obtained from several individual measurements in different combinations of O2, CO, and H2. The fact that the experimental points are sitting on the same curve combining the individual conduction mechanisms indicates that the reactions involving these gases have very similar effects. The upward band bending is determining a conduction mechanism dominated by the surface depletion layer; the downward band bending changes the conduction mechanism to one dominated by accumulated electrons in the surface layer. In the latter case, the deeper the conduction band edge on the surface falls below the Fermi level (higher concentrations of CO and H2 in the absence of oxygen), the weaker the effect of the band bending on the sensor signal becomes. The switch from one conduction model to the other takes place directly in a dry nitrogen atmosphere; this indicates that, under these conditions, a flat band situation is present (absence of active intrinsic surface traps). The significance of the results presented is not only limited to the conditions used here. In higher oxygen backgrounds (synthetic air), one could also find resistance changes on exposure to reducing gases (high concentrations) of a few orders of magnitude, which could result in a switch between the different conduction mechanisms. This could explain why it is sometimes so difficult to describe the dependence of the full sensor response on the target gas concentration over a large concentration range with a single curve. In the following chapter, experiments in more realistic conditions were used to further examine such a possible switch of the conduction mechanism.28 66 N. Bârsan et al.

2.5 Conduction mechanism switch for n-type SnO2ebased sensors during operation in application- relevant conditions

Again for this set of experiments, similar SnO2 sensors as described in Section 2.4.1 were used at 300C and exposed to the following conditions: N2; dry air; 0.3e250 ppm CO in dry air; humid air with 6% r.h. @25 C; 0.3e250 ppm CO in humid air with 6% r.h. @25C; humid air with 20% r.h. @25C; humid air with 50% r.h. @25C, 0.3e250 ppm CO in humid air with 50% r.h. @25C. Fig. 2.19 represents the time dependence of the resistance during exposure to different humidity levels and CO. The baseline resistance in pure N2drepresenting the ﬂat band conditiondis indicated by a dotted line. One can observe that during exposure to CO in humid air the resistance decreases much below the value corresponding to the absence of oxygen (N2 background). This means a change in the conduction mechanism from the one controlled by the depletion layer to the one controlled by the accumulation layer. The consequences are signiﬁcant for both sensing components: transduction and reception. In the case of the transduction, it means that the conductance is no more proportional to the surface concentration of free charge carriers but to the average one over the whole accumulation layer. This means that the same change in the band bending will result in a reduced conductance change

1000 Depletion layer Air Air 1M

] 100 Ω Air 6% r.h. CO in dry air 100k Air 20% r.h. Air 50% r.h. CO in air 6% r.h. CO in air 20% r.h. 10 CO in air 50% r.h. 2 N

Resistance [ 10k Accumulation layer 1

1k Concentration CO [ppm] Undoped SnO2 0,1 0,0 0 20 40 60 80 100 120 140 160 180 200 010203040 50 Time [h] Time [h]

Figure 2.19 DC electrical resistance measurement of an undoped SnO2 sensor (polycrystalline thick film sensing layer) during exposure to N2, and 0e250 ppm CO in different background conditions. Right: the profile of CO exposure as a function of time, which is valid for all background conditions. Conduction mechanism in semiconducting metal oxide sensing films: impact on transduction 67 and hence a lower sensor signal in case of an accumulation layer (see Eq. 2.18).

2.6 Conclusion and future trends The contribution presented here highlights the importance of the conduction mechanism in the SMOX sensing layers for the performance of the corresponding gas sensors. It basically demonstrates that high surface reactivity and the considerable charge transfer processes associated with it are not sufficient for “large” sensor signals. These depend to a large extent on the way in which the surface changes are translated into measurable changes of the electrical resistance of the sensor, which depend on the conduction mechanism. The proposed conduction models, which are based on simple assumptions and confirmed by the experimental results, explain the weaker performance of the devices based on p-type materials when compared with those based on n-type materials. They also open up new opportunities for investigation in combination with working condition characterization techniques. Future work will concentrate on applying the models for more complicated and realistic operational conditions in the direction indicated by the CuO investigation presented in Section 2.4.2. The understanding of the effect of the presence of humidity in the ambient atmosphere as well as the understanding of the effect of surface dopants and bulk doping are of crucial importance. References [1] Bârsan N, Koziej D, Weimar U. Metal oxide-based gas sensor research: how to? Sensor Actuator B Chem 2007;121(1):18e35. https://doi.org/10.1016/j.snb.2006.09.047. [2] Brattain W, Bardeen J. Surface properties of germanium. Bell Telephone Syst Tech Publ Monogr 1953;2086:1e41. [3] Bielanski A, Deren J, Haber J. Electric conductivity and catalytic activity of semiconducting oxide catalysts. Nature 1957;179(4561):668e9. https://doi.org/10.1038/ 179668a0. [4] Heiland G. Zum Einfluss von Wasserstoff auf die elektrische Leitf€ahigkeit von ZnO- Kristallen. Z Phys 1954;138:459e64. https://doi.org/10.1007/BF01327362. [5] Seiyama T, Kato A, Fujiishi K, Nagatani M. A new detector for gaseous components using semiconductive thin films. Anal Chem 1962;34:1502f. https://doi.org/10.1021/ ac60191a001. [6] Taguchi N. US patent No. 3631436. 1971. [7] Henrich VE, Cox PA. The surface science of metal oxides. Cambridge University Press; 1994. 0e521e44389-X. 68 N. Bârsan et al.

[8] Bârsan N. Conduction models in gas-sensing SnO2 layers: grain-size effects and ambient atmosphere influence. Sensor Actuator B Chem 1994;17(3):241e6. https:// doi.org/10.1016/0925e4005(93)00873-W. [9] Bârsan N, Weimar U. Conduction model of metal oxide gas sensors. J Electroceram 2001;7(3):143e67. https://doi.org/10.1023/A:1014405811371. [10] Dutraive MS, Lalauze R, Pijolat C. Sintering catalytic effects and defect chemistry in polycrystalline tin dioxide. Sensor Actuator B Chem 1995;26(1e3):38e44. https:// doi.org/10.1016/0925-4005(94)01552-S. [11] Weimar U, Morante JR, Schweizer-Berberich M, Bârsan N, Goepel W. Electrode effects on gas sensing properties of nanocrystalline SnO2 gas sensors. In: Conference proceedings EUROSENSORS XI, Warschau (Poland); 1997. ISBN 83-908335-0-6, 1377e80. [12] Oprea A, Bârsan N, Weimar U. Work function changes in gas sensitive materials: fundamentals and applications. Sensor Actuator B Chem 2009;142(2):470e93. https:// doi.org/10.1016/j.snb.2009.06.043. [13] Bârsan N, Weimar U. Understanding the fundamental principles of metal oxide based gas sensors; the example of CO sensing with SnO2 sensors in the presence of humidity. J Phys Condens Matter 2003;15(20):R813e39. PII S0953e8984(03)33587e8. [14] Ihokura K, Watson J. Stannic oxide gas sensors. In: Principles and applications. Boca Raton: CRC Press; 1994. [15] Williams DE. Semiconducting oxides as gas-sensitive resistors. Sensor Actuator B Chem 1999;57(1e3):1e16. https://doi.org/10.1016/S0925e4005(99)00133e1. [16] Kim YS, Hwang IS, Kim SJ, Lee CY, Lee JH. CuO nanowire gas sensors for air quality control in automotive cabin. Sensor Actuator B Chem 2008;135(1):298e303. [17] Li Y, Liang J, Tao Z, Chen J. CuO particles and plates: synthesis and gas-sensor application. Mater Res Bull 2008;43(8e9):2380e5. https://doi.org/10.1016/ j.materresbull.2007.07.045. [18] Miremadi BK, Singh RC, Chen Z, Morrison SR, Colbow K. Chromium oxide gas sensors for the detection of hydrogen, oxygen and nitrogen oxide. Sensor Actuator B Chem 1994;21(1):1e4. https://doi.org/10.1016/0925e4005(93)01208-L. [19] Shimizu Y, Nakashima N, Hyodo T, Egashira M. NOx sensing properties of varistor- type gas sensors consisting of micro p-n junctions. J Electroceram 2001;6:209e17. https://doi.org/10.1023/A:1011448513611. [20] Zhang J, Liu J, Peng Q, Wang X, Li Y. Nearly monodisperse Cu2O and CuO nano- spheres: preparation and applications for sensitive gas sensors. Chem Mater 2006;18(4): 867e71. https://doi.org/10.1021/cm052256f. [21] Bârsan N, Simion C, Heine T, Pokhrel S, Weimar U. Modeling of sensing and transduction for p-type semiconducting metal oxide based gas sensors. J Electroceram 2010; 25(1):11e9. https://doi.org/10.1007/s10832e009e9583-x. [22] Pokhrel S, Simion CE, Quemener V, Bârsan N, Weimar U. Investigations of conduction mechanism in Cr2O3 gas sensing thick films by ac impedance spectroscopy and work function changes measurements. Sensor Actuator B Chem 2008;133(1):78e83. https://doi.org/10.1016/j.snb.2008.01.054. [23] Huebner M, Simion CE, Tomescu-Stanoiu A, Pokhrel S, Bârsan N, Weimar U. In- fluence of humidity on CO sensing with p-type CuO thick film gas sensors. Sensor Actuator B Chem 2011a;153:347e53. https://doi.org/10.1016/j.snb.2010.10.046. [24] Morrison SR. The chemical physics of surfaces. New York: Plenum; 1977. ISBN 0e306e30960e2. [25] Huebner M, Pavelko RG, Bârsan N, Weimar U. Influence of oxygen backgrounds on hydrogen sensing with SnO2 nanomaterials. Sensor Actuator B Chem 2011b;154(2): 264e9. https://doi.org/10.1016/j.snb.2010.01.049. Conduction mechanism in semiconducting metal oxide sensing films: impact on transduction 69

[26] Bârsan N, Huebner M, Weimar U. Conduction mechanisms in SnO2 based polycrystalline thick film gas sensors exposed to CO and H2 in different oxygen backgrounds. Sensor Actuator B Chem 2011;157(2):510e7. https://doi.org/10.1016/ j.snb.2011.05.011. [27] Diéguez A, Romano-Rodríguez A, Morante JR, Kappler J, Bârsan N, Goepel W. Nanoparticle engineering for gas sensor optimisation: improved sol-gel fabricated nanocrystalline SnO2 thick film gas sensor for NO2 detection by calcination, catalytic metal introduction and grinding treatments. Sensor Actuator B Chem 1999;60(2e3): 125e37. https://doi.org/10.1016/S0925e4005(99)00258e0. [28] Bârsan N, Rebholz J, Weimar U. Conduction mechanism switch for SnO2 base sensors during operation in application relevant conditions; implications for modeling of sensing. Sensor Actuator B Chem 2015;207:455e9. This page intentionally left blank CHAPTER THREE

The effect of electrode-oxide interfaces in gas sensor operation

Sung Pil Lee1, Chowdhury Shaestagir2 1Kyungnam University, Changwon, Kyungnam, Korea 2Intel Corporation, Hillsboro, OR, United States

Contents

3.1 Introduction 72 3.2 Electrode materials for semiconductor gas sensors 74 3.2.1 Metals and conduction 74 3.2.2 Influence of electrode materials 77 3.2.2.1 Silver 83 3.2.2.2 Gold 83 3.2.2.3 Platinum 84 3.2.2.4 Palladiumesilver 84 3.2.2.5 Platinumesilver 84 3.2.2.6 Platinumegold 84 3.2.2.7 Palladiumegold 85 3.2.3 Electrode configuration 85 3.2.4 Electrode geometry 90 3.3 Electrode-oxide semiconductor interfaces 95 3.3.1 Ideal contact of metal and oxide semiconductor 95 3.3.2 Contacts with surface states and an interfacial layer 99 3.3.3 Image force effects on the barrier height 103 3.4 Charge carrier transport in the electrode-oxide semiconductor interfaces 104 3.4.1 Electric field and capacitance in the metal-semiconductor interface 104 3.4.2 Transport mechanism across the junction barrier 109 3.4.3 Tunneling effects in the oxide-semiconductor interface 112 3.4.4 Structure of the interfacial layer 115 3.4.4.1 State 0: the clean semiconductor surface 117 3.4.4.2 Stage 1: the dilute limit 118 3.4.4.3 Stage 2: monolayer formation 118 3.4.4.4 Stage 3: addition monolayers and interdiffusion 118

Semiconductor Gas Sensors, Second Edition ISBN: 978-0-08-102559-8 © 2020 Elsevier Ltd. https://doi.org/10.1016/B978-0-08-102559-8.00003-3 All rights reserved. 71 j 72 Sung Pil Lee and Chowdhury Shaestagir

3.5 Gas/solid interactions in the electrode-oxide semiconductor interfaces 119 3.5.1 Dipole formation in the interfacial layer 119 3.5.2 Effects of hydrogen adsorption in the Schottky barrier junction 120 3.5.3 Adsorption of other gases in the Schottky barrier junction 122 3.6 Conclusions 124 References 125

3.1 Introduction There is a growing demand for gas sensors for efficient use of energy and raw materials, as well as to reduce environmental pollution despite increasingly complex manufacturing processes. Taguchi sensor is the most well-known gas sensor1,2 that detects reducing gases, whereas an oxygen sensor based on an ion-conducting sensor is the second most famous e type.3 6 Research and development for gas sensors is conducted in two stages. The first stage is to develop a new sensor whose application is empirically optimized. The characteristics of sensitivity, selectivity, long-term drift, and reliability are defined, although its operation mechanism is not e fully understood.7 11 The second stage is to modify, optimize, and e standardize the system of the developed sensor.12 15 The developed sensors can measure change or values of current,16,17 impedance,18,19 capacitance,20 frequency,21 potential difference,22,23 and electromotive force.24 In addition, the correlation between the sensor structure and electrode is very important to expressively depict these phenomenological parameters that characterize the sensor. The semiconductor gas sensor is not an energy conversion (generator- type) but energy control (modulator-type) sensor (see, e.g., Fraden, Hand- book of Modern Sensors). The physical properties of a sensing material change on exposure to gas molecules, and external electric energy transmits the change as a sensor signal. This implies that, in most cases, the electrode of the semiconductor gas sensor is similar to that of an electronic device, which delivers current flow or electric power supply without loss or supplies electric energy from external power sources to the device. Thus, in conventional electronic devices, the electrode only connects the device and external circuit. Accordingly, a strong mechanical adhesion and small contact resistance are the most significant factors; in addition, durability, chemical e resistance, reliability, and cost should be considered.25 28 However, the electrode of a semiconductor gas sensor not only measures the electric The effect of electrode-oxide interfaces in gas sensor operation 73 properties of the sensor but also measures the catalytic properties of the sensing material. The ohmic electric contact made between the device and the electrode material is acceptable; however, the semiconductor gas sensor sometimes requires a rectifying contact between the sensing material and electrode. A rectifying contact would create a dipole in the interfacial zone of a metal and semiconductor triggered by gas adsorption, reducing a potential barrier from time to time or leading to complex phenomena such as field e emission or tunneling effect due to thermionic field emission.29 35 In special cases where the semiconductor gas sensor is applied to cars or in the aerospace industry, the electrode material should be able to operate above 600C.6,36 The surface and interface science for semiconductor gas sensors have been extensively studied since Seiyama et al.1 reported that the charge carriers in the surface of oxide semiconductor in contact with a gas varied e according to the gas concentration. In addition, gas sensing mechanism,37 39 gas sensor technology,40 the semiconductor junction for gas sensors,41 practical hydrogen sensors,42 and gas sensor design43 have been reviewed by several researchers. The electrode materials and geometry have advanced considerably in the last few decades. The physics of the energy barrier in an electrodee semiconductor interface could be significantly compared with the energy barrier in the contact of a doped semiconductor. Charge transfer during the chemical reaction of gas in the electrode/semiconductor interfaces leads to a uniform Fermi level instead of energy band bending. This chemical reaction in the interfaces would affect the conductance of the sensor, as well as chemical reaction in the semiconductor surface. To design reliable semiconductor gas sensors requires the understanding of electrodee semiconductor interfaces and control of the geometric and electronic structures of electrodes. The aim of this chapter is to describe and review the interface chemistry and transition theory of the electrode-oxide semiconductor layer in gas sensor operation. Section 3.2 deals with criteria for selecting the metal and semiconductor materials used in the fabrication of gas sensor. The chemistry and physics of barrier formation in the metal-oxide semiconductor interfacial layer are outlined in Section 3.3. The recent investigations into the charge carrier transport model, including the tunnel effect in the electrode-oxide semiconductor interface, are discussed in Section 3.4. Section 3.5 surveys research and development works that have been undertaken on the gas/solid interactions in the electrodeesemiconductor interfaces. 74 Sung Pil Lee and Chowdhury Shaestagir

3.2 Electrode materials for semiconductor gas sensors 3.2.1 Metals and conduction Understanding of the behavior of electrons in solids is one of the keys to understanding electrode materials. The electron theory of solids is capable of explaining electrical, optical, magnetic, thermal, and chemical properties of materials. In other words, electron theory provides important fundamentals for a technology which is often considered to be the basis for modern civilization. Electrical conduction involves the motion of charges in a material under the influence of an applied electric field. A material can be generally classified as a conductor if it contains a large number of free electrons or mobile charges carriers. In metals, due to the nature of metallic bonding, the valence electrons from the atoms form a sea of electrons that are free to move within the metal and are therefore called “conduction electrons.” This is especially true for pure metals, where atom size and pack- ing are uniform and nothing is present to dissipate the free motion of electrons. Alloying disrupts the uniformity of the structure and reduces the electrical conductivity. An increase in temperature also disrupts the structure because of lattice vibration and results in a decrease in electrical conductivity. Good electrical conductors, such as metals, are also known to be good thermal conductors. The conduction of thermal energy from higher to lower temperature regions in a metal involves the conduction electrons carrying the energy. Consequently, there is an innate relationship between the electrical and thermal conductivities, which is supported by theory and experiments.44 The conductivity, s, of different materials spans about 25 orders of magnitude, as shown in Fig. 3.1. This is a largest-known variation in a physical property. It is generally accepted that, in metals and alloys, the electronsdparticularly the outer or valence electronsdplay an important role in electrical conduction. Before making use of the electron theory, we need to remind of some fundamental equations of physics pertaining

Porcelain Fe Quartz Rubber Glass Si Ge Mn Ag NaCl Mica GaAs Cu σ [1/Ω·m] 10–18 10–16 10–14 10–12 10–10 10–8 10–6 10–4 10–2 1 102 104 106 108

Insulators Semiconductors Metals

Figure 3.1 Room temperature conductivity of various materials. The effect of electrode-oxide interfaces in gas sensor operation 75 to electrical conduction. These laws have been extracted from experimental observations.45 Ohm’s law V ¼ R$I (3.1) relates the potential difference V (in volts) with the electrical resistance R (in ohms) and the electrical current I (in amperes). A differential form of Ohm’s law is J ¼ s$E (3.2) which links current density, J ¼ I/Adi.e., the current per unit area (A/m2), with conductivity s and electric ﬁeld strength V E ¼ (3.3) L The resistance of a conductor can be calculated from its physical dimensions by Lr R ¼ (3.4) A where L is the length of the conductor, A is its cross-sectional area, and r is the speciﬁc resistance or resistivity. The conductivity is in inverse proportion to the resistivity:

r ¼ 1 s (3.5) Fig. 3.2 shows the net flow of electrons in a conductor cross-sectional area A in the presence of an applied field Ex. Notice that the direction of electron motion is opposite to that of the electric field Ex and of conventional current because the electrons experience a Coulombic force eEx,in the x direction, because of their negative charge. We know that the conduction electrons are actually moving around randomly in the metal, but we

Δx A - - - - ν - - - - dx ------Ix ------Figure 3.2 Drift of electrons in a conductor in the presence of an applied electric ﬁeld. 76 Sung Pil Lee and Chowdhury Shaestagir

will assume that, as a result of the application of the electric field Ex, they all acquire a net velocity in the x direction. Otherwise, there would be no net flow of charge through area A. The average velocity of the electrons in the x direction at time t is denoted as vdx(t). This is called the “drift velocity,” which is the instantaneous velocity vx in the x direction averaged over many electrons, (w1028 m 3); that is 1 v ¼ ½v þ v þ v þ ,,,þ v (3.6) dx N x1 x2 x3 xN where vxi is the x direction velocity of ith electron and N is the number of conduction electrons in the metal. Suppose that n is the number of electrons per unit volume in the conductor (n ¼ N/V). In time Dt, electrons move a distance Dx ¼ vdxDt, so the total charge Dq crossing the area A is enADx. This is valid because all the electrons within distance Dx pass through A; thus, n(AvdxDt) is the total number of electrons crossing A in time Dt. The current density in the x direction is Dq enAv Dt J ¼ ¼ dx ¼ env (3.7) x ADt ADt dx This general equation relates Jx to the average velocity vdx of the electrons. It must be appreciated that the average velocity at one time may not be the same as at another time because the applied field, for example, may be changing: Ex ¼ Ex(t).

JxðtÞ¼envdxðtÞ (3.8) To relate the current density Jx to the electric field Ex, we must examine the effect of the electric field on the motion of the electrons in the conductor. To do so, we will consider copper crystal. The copper atom has a single valence electron in its 4s subshell, and this electron is loosely þ bound. The solid metal consists of positive ion cores, Cu , at regular sites in the face-centered cubic crystal structure.2 The valence electrons detach themselves from their parent atoms and wander around freely in the solid, forming a kind of electron cloud or gas. These mobile electrons are free to respond to an applied field, creating a current density Jx. The valence electrons in the electron gas are therefore conduction electrons. If the electric field strength is not very high, then the drift velocity of conduction electrons is proportional to the electric field strength, vdxðtÞ¼mExðtÞ (m e mobility), and the Ohm’s law (3.2) follows from Eq. (3.8) with the conductivity given by s ¼ enm. The effect of electrode-oxide interfaces in gas sensor operation 77

3.2.2 Influence of electrode materials Many researchers have long studied the interaction between the electrode and sensor materials, as well as the impact of the electrode materials on e the sensing behavior.46 59 The types of electrode materials used for semiconductor gas sensors are classified into bulk, thick film, and thin film. The bulk type is rarely used for the semiconductor gas sensor. The thick film type, made by screen printing from a paste, and the thin film type, made by vacuum deposition, are employed in many cases. The impact of the electrode on the properties of gas sensors based on tin oxide has been studied mainly by comparing various electrode materials such as Au, Pt, e Ag, and Pd.29,46 54 Toohey60 summarized the study on the interaction between electrodes and sensor materials and the influence of electrode materials in sensing behavior. The most common electrode material in practical and experimental sensors appears to be platinum, although gold and silver are occasionally used. Chemically, Pt and Au are relatively inert. As pure metals, they can be sputtered or evaporation coated onto a substrate, and both are available in ink formulations for screen printing. Ball or wedge bonding with platinum, gold, or aluminum wire allows the sensor device to be packaged in a conventional semiconductor header. A study of SnO2 and SnO2eMn2O3 hydrogen sensors with gold, palladium, and platinum electrodes showed that changing from platinum to gold could produce many-fold increase in sensitivity and a shift in peak sensitivity temperature from w375 to w450C(Fig. 3.3). Also, while the pure tin dioxide sensors had linear IeV characteristics under all conditions, the mixed oxide devices showed nonlinearity for high hydrogen concentrations with palladium or gold electrodes, but did not so when platinum was used. This suggests that the electrodee sensor contact was an appreciable component of the total sensor impedance.47

140

120 Au 100 Pt

80 Pd Rg / 2

N 60 R 40

200 250 300 350 400 450 500 550 T ( °C)

Figure 3.3 Sensitivity inﬂuence of three electrodes on a SnO2eMn2O3 (10:1) sensor for hydrogen gas. 78 Sung Pil Lee and Chowdhury Shaestagir

Capone et al.46 analyzed the impacts of two different interdigitated electrode geometries on the sensitivity of two different electrode materials (Au and Pt) for CO gas. These studies revealed that the Au electrode had a lower stability level than the Pt electrode. With regard to temperature, the sensitivity of CO was the highest at approximately 300C for the Au electrode and at 450C for Pt electrode, and it was observed to decrease slightly at low temperatures. In addition, a pure tin oxide sensor has displayed linear currentevoltage properties under all conditions, whereas a sensor with an additive has shown nonlinear properties. The Pd and Au electrodes had nonlinear characteristics, but the Pt electrode had linear characteristics for high hydrogen concentrations. These studies reported that the electrodeesemiconductor contact exerts substantial influence on the entire sensor impedance.47 Saukko et al.48 studied the influence of electrode materials on the properties of a tin oxideebased gas sensor. The energy barrier between the electrode and the sensing semiconductor could be significant compared to the energy barriers between the semiconductor grains. Then, chemical reactions between the gas atmosphere and metalesemiconductor interface would strongly affect the overall conductance of the sensor. When SnO2 thick film gas sensors that use Au and Pt as electrode materials were tested for hydrogen and CO gas, it was observed that the Pt electrode was more sensitive to H2, whereas the Au electrode was more sensitive to CO. Durrani49 used Ag, Al, Au, and Pt to study the effect of electrode material on the SnO2-based CO thin film gas sensor. Pt and Au showed higher response than Ag or Al when the electrode material was below the sensing material. In addition, Gourari et al.,47 Pijolat,50 and Bertrand et al.51 have studied Pt and Au as electrode materials in SnO2 gas sensors. Schottky-type sensors, in which the metal and semiconductor are in contact, are most widely used as hydrogen sensors. When the gas is not adsorbed in Schottky-type sensors, the energy band of the semiconductor bends upward or downward by the difference in the Fermi level between the metal and the semiconductor in the thermal equilibrium state. Such a situation arises when there is a thin insulator layer between the metal and semiconductor as well.28,42 In general, Pd is used as the electrode material for Schottky-type H2 sensors. When hydrogen molecules adsorb onto Pd, which is a catalytic metal, they dissociate into hydrogen ions. Some of these ions permeate Pd, spread toward the metalesemiconductor interface, form dipoles, and then change the metal’s work function and, hence, the Schottky barrier height. This change in the Schottky barrier height causes a shift in the The effect of electrode-oxide interfaces in gas sensor operation 79 currentevoltage (IeV) characteristics and thus the response can be measured as the change in voltage when the diode is operated at constant bias current. In metaleinsulatoresemiconductor field effect transistor (MISFET)etype hydrogen sensors, the threshold voltage in the gate layer changes based on the hydrogen concentration, resulting in a change in the drain current. Hydrogen sensors that use Schottky diodes were proposed for the first time by Lund- strom et al.61 and Steele and MacIver59 Both diodes used palladium as the metal, and the semiconductor substrates used were n-Si and CdS, respectively. In 1979, Ito62 had predicted that Schottky diodes consisting of similar metals and ionic semiconductors (such as SnO2,In2O3,KTaO3,ZnO,etc.) would also be sensitive to hydrogen. Comparative studies between Schottky diodes using Pd and Pt as the catalytic metals indicated the superior performance of Pt in terms of speed of response and sensitivity to hydrogen.63,64 In addition to Pd and Pt, other hydrogen-sensitive metals and alloys had been proposed including Ru,65 Ni,66 Au,67 Ag,68 IrPt, and PdAg.69 Song et al.69 tested the response to hydrogen gas of AlGaN/GaN Schottky diodes with Pt, IrPt, and PdAg from 200 to 800C. From 200 to 300C, PdAg diodes exhibited significantly higher sensitivity compared with Pt and IrPt diodes. Above 400C, however, IrPt and Pt diodes showed higher sensitivity, while the sensitivity of PdAg diodes degraded because of the poor thermal stability (Fig. 3.4). Studies on the electrode effects of semiconductor gas sensors are summarized in Table 3.1. One of the main disadvantages regarding the metal oxideebased gas sensors is a gradual loss of stability and reliability: the problems of aging and drift of the sensors. Important factors in selecting an electrode material

4 Pt IrPt 3 PdAg

2 Sensitivity (S) 1

0 200400 600 800 Temperature (°C) Figure 3.4 The comparison of hydrogen sensitivity in AlGaN/GaN Schottky diodes with different catalytic metals. 80 Sung Pil Lee and Chowdhury Shaestagir

Table 3.1 Studies for electrode effects of semiconductor gas sensors. Electrode materials Sensing materials Target gases References

47 Au, Pd, Pt SnO2 H2 Gourari et al. 62 Pd, Pt ZnO, SnO2,In2O3, H2 Ito KTaO3 65 Ru SiC H2 Basu et al. 66 Ni Si H2 Salehi and Nazerian 67 Au ZnO H2 Pandis et al. 69 Pt, IrPt, PdAg Al GaN-GaN H2 Song et al. 48 Au, Pt SnO2 H2, CO Saukko et al. 53 Au, Pt SnO2 H2, CO Rank et al. 46 Au, Pt SnO2 CO Capone et al. 49 Ag, Al, Au, Pt SnO2 CO Durrani 51 Au, Pt SnO2 CO Bertrand et al. 70 Au Fe2O3eIn2O3 CO Golovanov et al. 55 Ag, Au ZnO CO, NO2 Lin et al. 71 Au WO3 NO2 Tamaki et al. 72 Au SnO2 NO2 Shaalan et al. 50 Au, Pt SnO2 Benzene Pijolat 52 Pt, Au, PteAu SnO2 H2O Ylinampa et al. 56 Al WO3 Cl2 Bender et al. for a gas sensor include the long-term stability, heat resistance, chemical resistance, and adhesion to a substrate. Long-term investigations will determine the usability of the sensors. According to Meixner and Lampe73 the main reasons for inadequate long-term stability are the changes of the metal oxide and the metal electrode, instability of the wire contacts, and interaction with an unsuitable sensor casing. The degradation of contacts is mainly due to the diffusion at the electrode and oxide interface or the interaction of electrode with the surrounding atmosphere.74 As an electrode material for semiconductor gas sensors, Ag is stable in air and used over a wide temperature range. However, Ag has a low long-term stability disadvantage and the degradation of contacts. Ag can easily move or migrate at temperatures above 300C. Au is also one of the most popular electrode materials owing to its high electric conductivity and reliability. However, it has the disadvantage of easily diffusing into the substrate (especially silicon) at a relatively low temperature. On the other hand, Pt is the most stable electrode material, with little degradation. However, it is expensive and has poor substrate adhesion. To improve the adhesion to the substrate, a “glue layer” of Cr, Ti, or W is needed between the electrode and the substrate. For good adhesion, Hoefer The effect of electrode-oxide interfaces in gas sensor operation 81 et al.75 used Ta, whereas Michel et al.76 used TiN as the glue layer. Sozza et al.77 also reported that the Ti/Pt layer can prevent the rather fast degradation as compared to the Ti/Au layer or the Ti/Pd/Au layer. Capone et al.78 studied the influence of the aging of the Ti/Au interdigitated electrical contacts on the responses of pure and Ni-, Os-, Pt-, and Pd-doped SnO2 thin films. They found that the use of Ti/Pt electrical contacts, which were more stable than Ti/Au or Ti/Pd/Au structures, could reduce one of the possible causes of aging that produced the drift of the sensor responses. Some semiconductor gas sensors use a conductive polymer as the electrode material. Most organic polymers are electrically nonconductive, but conductive polymers can be produced by providing a channel for electrons to travel along polymer chains or to jump from chain to chain.79 Such conductive polymers include polyaniline, polyacetylene, polypyrrole, poly(p-phenylene), polythiophene, and poly(p-phenylenevinylene), among e which polyaniline (PANI) is the most widely used.80 82 Fig. 3.5 shows examples of conductive polymers. Polyaniline has received significant attention as it has a high electrical conductance of 103 S/cm and has been reported to have metallic properties. According to the synthesis method, polyaniline can be divided into the following states (Fig. 3.6): (i) completely oxidized state (PB: 1 y ¼ 0, quinoid); (ii) intermediate oxidation state (EB: 1 y ¼ 0.5); and (iii) completely deoxidized state (LB: 1 y ¼ 1, benzenoid). EB is generally easily produced using an oxidizer such as (NH4)2S2O8 to oxidize aniline directly in the presence of protonic acid. LB can be easily obtained by applying a reducing agent such as hydrazine hydrate to EB, whereas PB can be produced using an oxidant such as m-chloroperoxybenzoic acid.83,84

H N n

Polyaniline Polyacetylene

N H n Polypyrrole Poly(p-phenylene)

CH CH S n Polythiophene Poly(p-phenylene vinylene) Figure 3.5 Various conductive polymers. 82 Sung Pil Lee and Chowdhury Shaestagir

H H H H N N N N

Leuco-emeraldine Base (LB)

HH N NNN y 1–y Emeraldine Base (EB)

N N NN

Pernigraniline Base (PB) Figure 3.6 Polyaniline bases according to oxidation states.

Conductive polymers have a high conductance of approximately 103 S/ cm as compared to that of an ITO electrode. Thus, they can enable the production of thin films through spin coating, which is much more economical and convenient than evaporation or sputtering. However, conductive polymers have some disadvantages, such as a property change during the production of doped polymer composition, the use of a nonvolatile solvent (m-cresol), and their color. For realizing a flexible sensor system in future, the metal electrode materials must be replaced by organic materials. Among organic materials, monomolecular pentacene has the highest level of charge transfer.80 Pentacene, however, has a disadvantage in terms of its manufacturing process, such as it is impossible to affect vacuum evaporation. Polythiophene derivatives are used as conductive polymers to replace pentacene, and they have high electric field mobility; however, they show a relatively low on-and-off ratio. Polyaniline and polypyrrole also have a low on-and-off ratio, but the ratio can be enhanced because the conductivity level of nanostructured polyaniline can be more easily adjusted than the doping level as compared to polymers. The replacement of the electrode material, which is the core part of flexible devices, is very important. Many experts expect that when the replacement technology is accomplished, the development of flexible devices will reach the stage of completion. Polyaniline can also be applied to a variety of fields, such as the electrode material of sensors, insulation layer of O-TFT, and channel material of an electrical transport layer. Notably, for polyaniline, the electrical conduction can be much more easily controlled than for carbon nanotubes by adjusting the protonic acid doping levels and by other appropriate methods.80 The effect of electrode-oxide interfaces in gas sensor operation 83

Recently, the ﬁeld of printed electronics has been receiving considerable attention because of the development of semiconductor fabrication technology.85 It is important for semiconductor gas sensors to make electrodes by printing the materials. The most commonly used materials for electrodes are precious metals, such as silver, gold, platinum, and palladium. The alloys of these metals are also widely used.

3.2.2.1 Silver Silver inks were probably the first thick film inks to be developed and were used mainly in the construction of capacitors. There are several points in favor of silver as a conductive ink. Not only is it the least expensive metal which is compatible with the normal thick film process, but also it may be made to have good bond strength and high conductivity and is easily wetted by tin-lead solder. Although its leach resistance to this solder is poor, this is easily overcome by slight modification of the solder. Silver compositions are also compatible with several resistor and dielectric systems. The major disadvantage of silver is its strong tendency to migrate over the surface of insulants and resistors when subjected to electrical fields under conditions of high humidity; this may lead to the lowering of resistance or complete short circuits in other thick film components.

3.2.2.2 Gold Gold inks may be constituted to have high conductivity, similar to that of silver, and produce films which are stable under all normal service conditions. They are compatible with most dielectric and some resistor systems, although they are not generally suitable for terminating the palladium/silver type of resistors. The main disadvantages of gold are its high cost and its unsuitability for solder joining. Tin-lead solders are unsuitable for use with gold conductors and cannot easily be modified, as is the case of silver, although special solders such as gold-tin alloys may be used. Gold is normally used in circuits where high conductivity and reliability are required and in applications where silicon devices are to be eutectically bonded, or where ultrasonic bonds have to be made, to gold or aluminum wires. A further use of gold is in the closure of hermetic packages, where lids may be sealed to gold metallization using a solder alloy consisting of gold (80%)/tin (20%). The advantage of gold in this situation is that, in a neutral atmosphere, the solder seal may be made without the use of flux. 84 Sung Pil Lee and Chowdhury Shaestagir

3.2.2.3 Platinum Platinum inks are the most expensive of thick film conductive inks, but they are occasionally used where extreme resistance to molten solder and to bold strength degradation by solder is required. 3.2.2.4 Palladiumesilver Palladiumesilver alloys are perhaps the most widely used conductor compositions. They are less expensive than gold alloys, are compatible with most dielectric and resistor systems, and are suitable for ultrasonic wire bonding. The sheet resistivity is typically in the region 0.01e0.04 U/sq and, although this figure is considerably higher than the resistivity of pure metal conductors, it is lower than the figures for gold alloy conductors. The addition of palladium to silver greatly reduces the rate of dissolution of the metal in molten solder. Increasing the palladium content thus provides greater leach resistance, but at the expense of solderability and conductivity. It also increases the cost. It is common practice, therefore, for ink manufacturers to produce a range of palladiumesilver alloys of different palladium content so that the best compromise may be chosen for any particular application. Palladiumesilver pastes can be fired with excellent initial bond strength to the substrate, but this rapidly degrades if the circuits are stored at elevated temperature (above 70C) when the conductors are tinned. A further disadvantage is the possibility of silver migration under conditions of high humidity. The rate of migration is, however, considerably reduced by the presence of the palladium. 3.2.2.5 Platinumesilver Increasing world demand for palladium and wild fluctuations in its price have recently persuaded some ink manufacturers to add platinumesilver alloys to their range of conductors as alternatives to palladiumesilver. Sheet resistance ranges from 0.01 to 0.04 U/sq with increasing platinum content, and in this and most other respects the two ranges are found to be generally equivalent. Platinumesilvers are not, however, recommended for hybrid applications involving ultrasonic aluminum wire bonding.

3.2.2.6 Platinumegold Platinumegold systems possess many of the advantages of both gold and platinum. They have excellent solderability combined with outstanding resistance to solder leaching and are also suitable for both wire and die bonding. They are compatible with most other thick ﬁlm materials and, while the initial bond strength tends to be lower than that of palladiumesilver alloys, they have The effect of electrode-oxide interfaces in gas sensor operation 85 much greater resistance to solder bond strength degradation. The chief disadvantages lie in their high cost and rather high electrical resistivity (0.08e0.1 U/sq).

3.2.2.7 Palladiumegold This alloy system has generally similar properties to platinumegold and is less costly. The solder leach resistance and solder aging are, however, inferior to the more expensive material. Resistivity is of the order 0.04e0.10 U/sq. Table 3.2 summarizes the properties of metal inks for semiconductor gas sensors. 3.2.3 Electrode configuration The electrodes used for gas sensors should be in contact with the substrates, and their electrical properties should be easily measured. The following conditions are thus required: (1) They should be chemically and mechanically stable on the substrates. (2) The connection to the lead-out terminals should be easy. (3) The sensing film should not be damaged during electrode formation. (4) They must have a geometry that is suitable for sensor construction. The two-electrode type configuration, in which the gas sensing material is positioned between two metal electrodes, is the most widely applied one in semiconductor gas sensors. Occasionally, a third electrode is used as a heater for the sensors. Toohey60 explained the electrode types used in semiconductor gas sensors. Two-electrode configurations are used for gas sensors, as shown in Fig. 3.7. In type (a), the Pt electrode is formed on an alumina cylinder, which is applied to a Figaro sensor, and then the sensing materials are deposited on it and sintered. In type (b), a tablet made of an oxide semiconductor is sintered and then the electrode is formed on both sides. In type (c), two combs face each other to create an interdigitated geometry on the substrate. The transmission type (d) sensor is formed, to fabricate a surface acoustic wave filter that measures the frequency changes. The interdigitated geometry is the most widely accepted geometry for the electrodes of a gas sensor, as it enables a wide contact area between the electrodes within the limited area. In addition, it forms the electrodes first and then deposits the sensing materials on them, thereby causing no damage to the sensing materials. A one-electrode configuration that differs from the two-electrode type that has been previously used for semiconductor gas sensors has been developed. Korotcenkov43 reviewed the design and type of the one-electrode 86 Sung Pil Lee and Chowdhury Shaestagir

Table 3.2 Properties of printing metals and alloys for electrode of semiconductor gas sensors. Materials Electrical properties Advantages Disadvantages

Silver - High conductivity - Least expensive - Tendency to - Compatible with - Good bond strength migrate over the resistor and surface of insulants dielectric system and resistors under - Resistivity: high humidity 1.59 ⅹ 10 8 U m Gold - High conductivity - Alloy with tin may - High cost and reliability be made without - Unsuitability for - Resistivity: the use of flux solder joining 2.44 ⅹ 10 8 U m Platinum - Use where extreme - Available wire, flat - Most expensive resistance to molten plate, and tube solder and to bond - Large range of size strength degradation - Useable in high by solder is required temperature - Resistivity: 11.0 ⅹ 10 8 U m Palladium - Compatible with - Suitable for - The possibility of esilver resistor and ultrasonic wire silver migration dielectric system bonding under high - Sheet resistance: humidity 0.01e0.04 U/sq Platinum - Alternatives to Pd - Not recommended esilver eAg for hybrid - Sheet resistance: applications 0.01e0.04 U/sq involving ultrasonic wire bonding Platinum - Compatible with - Excellent - High cost egold most thick film solderability - Rather high materials - Suitable for both electrical resistivity - Sheet resistance: wire and die 0.08e0.1 U/sq bonding Palladium - Similar properties to - Less expensive than - Inferior solder leach egold PteAu PteAu resistance and solder - Sheet resistance: aging than PteAu 0.04e0.10 U/sq configuration for semiconductor gas sensors. One electrode acts as both the heater and the measuring terminal, unlike the two electrode setup. As demonstrated in Fig. 3.8, one-electrode gas sensors can be formed by applying the metal oxide in the form of a bead on the electrode material The effect of electrode-oxide interfaces in gas sensor operation 87

(a) (b) (c) Wire Pt wire Sensing Sensor electrode Electrode Electrode material Heater Pt wire Heater Catalyst Catalyst Heater Alumina tube Wire Heater electrode

(d) (e) Sensing Sensing Contact pad material film

Output

Signal Substrate Interdifited electrode Interdigited Piezoelectric electrode substrate Figure 3.7 Two electrode conﬁguration used in gas sensors: (a) cylinder, (b) disk, (c) parallel plates, (d) interdigit, and (e) surface acoustic wave (SAW) line.

(a) (b) Pt heater Ceramic bead Sensing material (electrode) Electrode

Supporter Bonding pad

Lead wire Figure 3.8 One-electrode conﬁguration: (a) ceramic bead surrounding Pt electrode and (b) Pd electrode on alumina substrate. or by shunting the electrode through a coating. For the design of 86e89 one-electrode semiconductor sensors, materials such as SnO2, 90 70,91 In2O3, and Fe2O3 have mainly been used. Fig. 3.9 shows schematically a standard two-electrode sensor and one-electrode sensors in planar design. The schematic electrical circuits of these sensors are also shown in Fig. 3.9, where RPt is a coil resistance of Pt spiral, RMeO is interturn resistance of metal oxide ceramics, and RS is a total resistance of the sensor. For operation of the one-electrode sensors, impedance matching should be performed between the shunting semiconductor resistance and the electrode resistance, as shown in Fig. 3.9. One can adjust the electrode resistance by varying the electrode thickness, distance between the electrodes, or resistance of the sensing materials by adding additives to the oxide semiconductor or by modifying the thickness.43 88 Sung Pil Lee and Chowdhury Shaestagir

(a) (b) Conductive gas Conductive gas sensing metal oxide Pt contactsensing metal oxide Pt planar heater

Dielectric substrate Pt planar heater Dielectric substrate Capsulating layer R R R MeO S MeO I (const)

R R Pt Pt Standard solid state One-electrode conductometric sensor semiconductor sensor Figure 3.9 Planar constructions of standard (two-electrode) and one-electrode semiconductor sensors.

Faglia et al.92 used four-probe array analysis in the gas detection system to distinguish between the grain contribution and the contact contribution. This suggested that the contact contribution was very important for CO detection, while the material contributes to CH4 detection in tin oxide 93 gas sensors. For four-electrode semiconductor sensor design, CrTiO3 94 and WO3/TiO2 have mainly been used. It is clear that the material and geometry of the electrodes can inﬂuence gas sensor behavior. Many researchers are investigating the fabrication of gas sensors using nanoparticulate materials as the sensitive layer. While it is possible to use “normal”-sized electrodes with widths and separations of several microns for these devices, it is of interest to examine the changes in response which are obtained when nanoelectrodes are used; i.e., contacts of comparable dimensions to a single particle around 5 nm. Potential advantages of nanoelectrodes include the following60: (i) the possibility of addressing single nanodots; (ii) the ability to vary the relative contributions of electrodeedot and dotedot contacts to the total sensor resistance; (iii) where a nanodot ﬁlm consists of conducting and nonconducting particles, decreasing the electrode size could increase sensitivity by around an order of magnitude or more by “softening” the percolation threshold;95 (iv) small electrode systems use less sensor chip area; however, producing structures of these sizes is problematic. The effect of electrode-oxide interfaces in gas sensor operation 89

As all electronic parts become integrated and intelligent, it is also inevi- table to make small and integrated gas sensors. So far, many researchers have e used conventional MISFETs61,96 105 or microelectromechanical systems e (MEMSs)106 112 to manufacture semiconductor gas sensors. Gas detection with such technologies depends on the varying conductivity owing to gas adsorption and the reaction on the metaleinsulatoresemiconductor (MIS) structure surface or varying work functions of the MISFET resulting from catalytic reaction in the gate electrode. However, sensor stability is not ensured yet, although MIS gas sensors are increasingly needed. As the gate electrode is exposed, unintended reactions between the gate electrode and materials near it reduce the sensor sensitivity or selectivity with time and it takes longer to respond to the gas molecules. The gate electrode of the CO gas sensors with the MIS structure needs to apply a voltage for the device so as to form a channel and also carry out catalytic actions.105 Thus, the electrode can be made porous so that the area where the adsorbed gases contact the sensing materials increases instead of the gate covering the surface of the sensing materials. Janata and Josowic113 created a suspended microgrid on a FET gate to extend the lifetime of the sensing gas. In these devices, the gate metal is preceded by an additional space, which, in the case of GasFET, is permeable to gases. The suspended grid above the gate insulator is made of Pt or Au. Applying a Pd layer to this creates a hydrogen sensor. If a conductive polymer layer, such as polypyrrole, is deposited on the metal grid, then the sensor is sensitive to alcohols. In both cases, the reaction of the gas with the surface of the suspended metal grid or with the surface of the insulator causes a change in the electric field that is detected in the modified drain current.114 Lee et al.17,115 have tried to deposit a porous metal gate for humidity sensitive field effect transistors (HUSFETs) that can sense humidity. Here, a thin gold film of approximately 100 Å through which water molecules could penetrate was deposited on the active layer before a pattern was formed using lift-off techniques. When water molecules meet carbon nitride through the porous gold layer of the gate in Fig. 3.10, adsorbed water molecules on the carbon nitride are able to form dipole and to reorient freely under an applied gate voltage, resulting in an increase in the dielectric constant.17 Thereafter, Fukuda et al.105 applied porous Pt as gate electrode materials to improve the sensitivity of the MOSFET-type hydrogen sensors. When a porous electrode was used, the sensor detected 22 ppm of H2 gas in less than 2 min, thus indicating a remarkable gas detecting performance. Its sensitivity level was enhanced by approximately 10 times as compared to 90 Sung Pil Lee and Chowdhury Shaestagir

Gate pad Reference passivation

Source pad Mental 2 Mental 1

Drain pad

N+

N+ Locos

Porous Au N+ CNx Si3N4 N+ SiO2 Source Oxide Drain Oxide Sensor Oxide

Reference

Figure 3.10 Design of differential humidity sensitive field effect transistors with porous Au gate. that of a nonporous Pt surface because of the catalytic property of the porous Pt surface. For the purpose of detection of negative ions in the air, Lee et al.116 have suggested a nanoFET sensor that uses a TieAl layer as the electrode for the source and drain, while using a floated Ti/Au layer as the electrode on the gate oxide. 3.2.4 Electrode geometry Many researchers have studied the influence of the geometry and position of e electrodes on the sensitivity and selectivity of sensors.60,71,72,117 121 The width of digits in interdigitated electrodes or the space between the electrodes can affect the sensor performance. In other words, when the electrode spacing is narrow, the current between electrodes flows only in the film area right above it. On the contrary, when the spacing is wide, the current flows both horizontally and vertically throughout the film, thereby sampling a wider area.60,118 In addition, the electrodee semiconductor interface itself can cause a change in the device sensitive resistance. When the width/gap ratio of the electrode is changed, the influence of both the interface and the film resistance on sensitivity can be relatively reduced. The effect of electrode-oxide interfaces in gas sensor operation 91

Vilanova et al.119 studied the influence of electrode position, electrode gap, and active layer thickness for high, medium, and poor catalytic activity sensor/gas pairs. The purely geometric effect arises because the film conductance does not change instantly or uniformly when the gas ambient changes: the gas must diffuse through the film, reacting with the particle surfaces as it does so. A numerical simulation indicated, for example, that where a sensor is highly sensitive to the test gas, sensitivity increased with electrode spacing when the electrodes were underneath the film but decreased with spacing when the electrodes were deposited on top of the film. In contrast, when an electrode was placed above the sensing film, the sensor sensitivity decreases as the spacing between the electrodes increased. The result of injecting a highly reactive gas was same as the result of injecting a low reactive gas when the gap between the electrodes decreased and became smaller than the film’s thickness. Fig. 3.11 shows that the sensitivity depends on electrode spacing for a sensor whose electrode is placed below the sensor film. In this case, the detection level for even a highly reactive gas was observed to be the same as that of a low-reactive gas. On the contrary, when the electrode gap is sufficiently wide, the detection level of even a low-reactive gas was observed to be the same as that of a highly reactive gas. Therefore, the gas detection performance of a sensor with an electrode placed below its film is better when the electrode width is wide and electrode spacing is narrow.

1E+04

1E+03

1E+02 0 G

/ 1E+01 G Δ

1E+00

1E–01 Bottom 1E–02 1E–01 1E+00 1E+01 1E+02 1E+03 W (μm) Figure 3.11 Sensitivity versus electrode gap for electrodes placed bottom. 92 Sung Pil Lee and Chowdhury Shaestagir

Gardner120 derived expressions which defined the response of a pair of planar conductometric gas sensors according to the electrode thickness and an electrode gap. The steady-state conductance in air, Go, of a homogeneous film of conductivity so and thickness L lying on semiinfinite electrodes can be found by integrating the current density over a closed surface; hence, 2 1=23 w2 1 þ 1 þ s b 6 2 7 ¼ 0 4 L 5 G0 p ln w (3.9) 2L where w is the separation of the electrodes and b is the length of the electrodes. It is assumed that the edge effects can be neglected (b [ w). When a gas of concentration C0 is introduced, it can diffuse into the porous film and react at sites dispersed uniformly throughout the film. These reaction sites modify the local conductivity of the film according to some function that depends on the local gas concentration Cx. The steady-state response R (fractional change in conductance) of the sensor by integrating the concentration-dependent current density can be given by GðC Þ G sðC Þ s R ¼ x o ¼ x o Gop so h i R 2 2 1=2 x=L¼1 ð Þ x þ w x = ¼ F Cx d ¼ x L("0 L 2#,L )L = (3.10) w2 1 2 w ln 1 þ 1 þ 4L2 2L

In the case of the narrow-gap sensor, the baseline conductance Gon and the response of the narrow-gap sensor Rn become sob 4L Gon ¼ ln p wn where wn/L 1 1 R = ¼ x 2 w 2 2 x x L 1 FðC Þ þ n d x=L¼0 x L 2L L R ¼ (3.11) n 4L ln wn The effect of electrode-oxide interfaces in gas sensor operation 93

In the case of the wide-gap sensor, the baseline conductance Gow can be simpliﬁed by using the ﬁrst term in Maclaurin expansion of the logarithmic function to give

2 sobL Gowz (3.12) p ww The sensing electrodes behave like a parallel-plate structure. As the electric field inside the film is nearly constant and independent of the distance x, the response of the wide-gap sensor Rw can be reduced to Z x=L¼1 x Rwz FðCxÞd (3.13) x=L¼0 L Under uniform gas profile (type I), the steady-state response function is given by the power law ð Þ¼ n < < FI Co k2Co 0 n 1 (3.14) where Co is the external gas concentration, k2 is the gas-sensitivity parameter for a semiconducting oxide material and exponent n normally lies between 0.3 and 0.9. The steady-state response of this type of sensor pair is as straightforward as the concentration profile (Fig. 3.12(a)). When a semiconducting oxide material is doped with catalyst of high activity (moving- boundary gas profile, type II), it is probable that the gas entering the film is

(a) (b) C Increasing value of 0 Air Active film Substrate

x x C Increasing value of 0 C C

Cp Gas concentration, Gas concentration, Air Active film Substrate

1.0 0.5 0.0–0.5 1.0 0.5 0.0 Distance in film, x/L Distance in film, x/L Figure 3.12 Gas profile in the active film: (a) uniform and (b) moving boundary. 94 Sung Pil Lee and Chowdhury Shaestagir rapidly consumed and so does not penetrate all the way into the film (Fig. 3.12(b)).120 The response function can be written as n k2C0 xb x L FII ðCxÞ¼ (3.15) 00 x < xb The response of the wide-gap sensor becomes simply the fraction of the film penetrated by the gas multiplied by the isotherm; hence, Z 1 z n x ¼ n xb Rw k2Co d k2Co 1 (3.16) xb=L L L The response of the narrow-gap sensor can also be found from Eqs. (3.11) and (3.15): ( ," #) 1=2 x x 2 w 2 ln 2 b þ b þ n L L2 4L2 R zk Cn (3.17) n 2 o 4L ln wn Tamaki et al. studied the effects of gap size differences between electrodes.71 Microgap electrodes (0.1e1.5 mm) were formed on the silicon substrate using the MEMS process, and WO3 films were deposited on these electrodes. When the gap size was larger than 0.8 mm, there was no change in sensitivity to NO2; however, when it was smaller than 0.8 mm, the sensitivity tended to increase and was expected to increase further in the range of less than 0.1 mm. They explained that by the number of grains of WO3 in the microgap and the resistance change at the boundaries. Shaalan et al.72 also fabricated microgap electrodes (1e30 mm) by dc sputtering and FIB techniques and deposited SnO2 nanowires on them using the suspension dropping method. They suggested that the interface between the electrodes and the sensing area plays a very important role in the sensing mechanism of SnO2 gas sensors. Comparison between the small gap and large gap electrodes showed that the small gap electrode had the advantage of reliability and high sensitivity to low NO2 concentration, whereas the large gap electrode had relatively high sensitivity for high concentrations. Hoefer et al.121,122 used an array of electrodes of differing width and separation to examine contact resistance effects in tin dioxide sensors. The transmission line method they used involves measuring the total resistance of a semiconductor sample as a function of electrode separation. It was shown that the modified sheet resistance displayed greater sensitivity to The effect of electrode-oxide interfaces in gas sensor operation 95

CO and NO2 than either the sheet resistance itself or the contact resistance. In this case, wide electrodes with narrow spacing would produce the most sensitive detection. An array of electrodes varying in width and spacing, but all using the same sensing material, could be used to resolve a mixture of CO, CH4,NO2, and water vapor into separate measurements of each component by ﬁrst determining the relative sensitivity of the total resistance of each electrode pair to the individual gases.123

3.3 Electrode-oxide semiconductor interfaces 3.3.1 Ideal contact of metal and oxide semiconductor The potential barrier which forms when a metal is contacted with an oxide semiconductor arises from the separation of charges at the metal-oxide semiconductor interface, such that a high-resistance region devoid of mobile carriers is created in the oxide semiconductor. This is similar with Schottkye Mott model that explains the barrier height of metal-semiconductor contact.124 According to this model, the barrier results from the difference in the work functions of the two substances. The energy band diagrams in Fig. 3.13 illustrate the process of barrier formation. Fig. 3.13(a) shows the electron energy band diagram of a metal of work function Fm(¼efm) and an n-type semiconductor of work function Fs(¼efs), which is smaller than Fm. The work function of a metal is defined as the amount of energy required to raise an electron from the Fermi level to the vacuum level. The vacuum level is the energy level of an electron just outside the metal with zero kinetic energy and is the reference level in Fig. 3.13(a). The work function Fm has a volume contribution due to the periodic potential of the crystal lattice and a surface contribution due to the possible existence of a dipole layer at the surface. The work function Fs of the semiconductor is defined similarly and is a variable quantity because the Fermi level in the semiconductor varies with the doping. An important surface parameter which does not depend on doping is the electron affinity c defined as the energy difference of an electron between the vacuum level and the lower edge of the conduction band. The work functions Fm and Fs and the electron affinity c are usually expressed in electron volts (eV). Note that the semiconductor shown in Fig. 3.13(a) does not contain any charges at the surface so that the band structure of the surface is the same as that of the bulk and there is no band bending. Fig. 3.13(b) shows the energy band diagram after the contact is made and equilibrium has been reached. When the two substances are brought into intimate contact, electrons 96 Sung Pil Lee and Chowdhury Shaestagir

(a) Vacuum level

eχ eΦs eΦm Ec EF EF EFi

Eν

(b)

eVbi eΦB0 Ec E F EF

eΦn

Eν Depletion region xn = W

Figure 3.13 Energy band diagram of metal contact to n-type semiconductor with Fm > Fs: (a) neutral materials separated from each other and (b) thermal equilibrium situation after the contact has been made, where xn is the penetration of the depletion region into the n material, W is the width of the depletion region, and EFi is intrinsic Fermi level. from the conduction band of the semiconductor, which have higher energy than the metal electrons, flow into the metal till the Fermi level on the two sides is brought into coincidence. As the electrons move out of the semiconductor into the metal, the free electron concentration in the semiconductor region near the boundary decreases. As the separation between the conduction band edge Ec and Fermi level EF increases with decreasing electron concentration and in thermal equilibrium EF remains constant throughout, the conduction band edge Ec bends up as shown in Fig. 3.13(b). The conduction band electrons which cross over into the metal leave a positive charge of ionized donors behind, so the semiconductor region near the metal gets depleted of mobile electrons. Thus, a positive charge is established on the semiconductor side of the interface and the electrons, which cross over into the metal form a thin sheet of negative charge contained within The effect of electrode-oxide interfaces in gas sensor operation 97 the ThomaseFermi screening distance from the interface (z0.5 Å). Consequently, an electric field is established from the semiconductor to metal in Fig. 3.13(b). Note that the width of space charge layer in the semiconductor is appreciable because the donor concentration in the semiconductor is several orders of magnitude smaller than the electron concentration in the metal.41 Let us now investigate how much the energy bands in the semiconductor will bend upward. It should be evident that because the band gap of the semiconductor is not changed by making contact with the metal, the valence band edge Ev will move up parallel to the conduction band edge Ec. Also, the vacuum level in the semiconductor will follow the same variations as Ec. This is because the electron affinity of the semiconductor is assumed to remain unchanged even after the metal contact is made. Thus, for metalesemiconductor system in thermal equilibrium, the important point which determines the barrier height is that the vacuum level must remain continuous across the transition region. Hence, the vacuum level from the semiconductor side must approach the vacuum level on the metal side gradually to preserve the continuity. The amount of band bending, then, is just equal to the difference between the two vacuum levels, which is equal to the difference between the two work functions. This difference is given by eVbi ¼ (Fm Fs), where Vbi is expressed in volts and is known as “contact potential difference” or the built-in potential of the junction: eVbi obviously is the potential barrier, in which an electron moving from the semiconductor into the metal has to surmount. However, the barrier looking from the metal toward the semiconductor is different and is given by124,125

FB0 ¼ðFm cÞ (3.18) As Fs ¼ c þ Fn, we have

eVbi ¼ FB0 Fn (3.19) where Fn ¼ (Ec EF) represents the penetration of the Fermi level in the band gap of the semiconductor and e is the electronic charge. Eq. (3.18) was stated by Schottky126 and, independently, by Mott.127 The exact shape of the potential barrier can be calculated from the charge distribution within the space charge layer. In most cases, the height fB0 of the potential barrier is orders of magnitude larger than the thermal voltage, kT/e, and the space charge region in the semiconductor becomes a high-resistivity depletion region devoid of mobile carriers. The shape of the barrier is then 98 Sung Pil Lee and Chowdhury Shaestagir determined from the donor distribution in the semiconductor. Schottky assumed the semiconductor to be uniformly doped up to the metal interface, which gives rise to a uniform charge density in the depletion region. The electric field strength for this constant space charge rises linearly with distance from the edge of the space charge layer and the resulting parabolic barrier is known as a “Schottky barrier.”126 Mott127 assumed a thin layer of semiconductor, devoid of any charge, sandwiched between a uniformly doped semiconductor and the metal. The electric field strength in thin region is constant and the potential increases linearly across this region. This type of barrier is known as the “Mott barrier.” The Mott barrier is encountered in situation where a thin layer of low-doped, nearly intrinsic semiconductor is interposed between a metal and a heavily doped semiconductor. The above description applies only to n-type semiconductor whose work function is less than the metal work function Fm. The electron energy band diagrams for an n-type semiconductor with Fm < Fs are shown in Fig. 3.14. Fig. 3.14(a) shows the energy bands for separated materials. After the contact is made, electrons flow from the metal into the conduction band

(a)

eΦs eΦm

E F Ec EF EFi

Eν

(b)

eΦBn eΦn

Ec E F EF

Eν

Figure 3.14 Energy band diagram of metal contact to n-type semiconductor with Fm > Fs: (a) neutral materials separated from each other and (b) contact under thermal equilibrium. The effect of electrode-oxide interfaces in gas sensor operation 99

Table 3.3 Work function of some important metals. Metal Work function (eV) Metal Work function (eV)

Pt 5.65 Zn 4.33 Ni 5.25 Al 4.28 Pd 5.12 Ag 4.26 Au 5.1 Pb 4.25 Cu 4.65 Ta 4.25 W 4.55 Cd 4.22 Cr 4.5 Ga 4.2 Hg 4.49 In 4.12 Sn 4.42 Zr 4.05 Ti 4.33 Cs 2.14 of the semiconductor, leaving behind a positive charge on the metal and causing an accumulation of electrons on the semiconductor side of the boundary. When equilibrium is reached, the Fermi level in the semiconductor is raised by an amount (Fs Fm) as shown in Fig. 3.14(b). The accumulation layer charge in the semiconductor is confined to a thickness of the order of Debye length and is, essentially, a surface charge. As the concentration of electrons in the metal is very large, the positive charge on the metal side is also a surface charge contained within a distance of about 0.5 Å from the metal-semiconductor interface. It is clear that no depletion region is formed in the semiconductor and there is no potential barrier for the electron flow either from the semiconductor toward the metal or in the opposite direction. The electron concentration is increased in the region near the interface and the highest resistivity region in the system is the bulk semiconductor region. The foregoing discussion has shown that, in case of n-type semiconductor, a metal-semiconductor contact is rectifying if Fm > Fs and is nonrectifying if Fm < Fs. The opposite is true for a metal p-type semiconductor contact. Table 3.3 gives the most preferred experimental values of e work function for important metals.125,127 133 3.3.2 Contacts with surface states and an interfacial layer The periodicity of the crystal lattice terminates at the surface of a semiconductor. In a crystal, the surface atoms have neighbors only on the semiconductor side; on the vacuum side, there are no neighbors with whom the surface atoms can make bonds. Thus, each of the surface atoms has one broken bond in which only one electron is present and the other is missing. The broken bonds are known as “dangling bonds.” Dangling bonds give rise 100 Sung Pil Lee and Chowdhury Shaestagir to localized energy states at the surface of the semiconductor with energy levels lying in the forbidden gap. These surface states are usually continuously distributed in the band gap and are characterized by a neutral level F0. The position of this neutral level is such that, when there is no band bending in the semiconductor, the states are occupied by electrons up to F0, making the surface electrically neutral. The states below F0 are donor-like because they are neutral when occupied and are positive when empty. Obviously, the states above F0 behave as acceptor-like. On clean surface of semiconductor, the density of surface states equals the density of surface atoms. Adsorbed layers of foreign atoms may considerably reduce this density by completing the broken bonds. The surface states modify the charge in the depletion region and, thus, affect the barrier height. Fig. 3.15(a) shows the electron energy band diagram of an n-type semiconductor under flat band condition. When a metal is now brought in contact with the semiconductor and equilibrium is reached, the Fermi level in the semiconductor must change by an amount equal to the contact potential by exchanging charge with the metal. If the density of surface states at the

(a)

Ec E EF F Surface state

Eν

(b)

Ec EF EF Surface state

Eν

Metal Semiconductor Figure. 3.15 Energy band diagram of the band bending interface in the presence of surface states for a metal/n-type semiconductor: (a) under ﬂat band condition and (b) after contact formation. The effect of electrode-oxide interfaces in gas sensor operation 101 semiconductor surface is very large, then the charge exchange takes place largely between the metal and the surface states, and the space charge in the semiconductor remains almost unaffected. As a result, the barrier height in Fig. 3.15(b) becomes independent of the metal work function. In this case, the barrier height is said to be pinned by surface states.134 Considering a continuum of interface states, a phenomenological formula for Schottky barrier height can be formulated.135,136

FB0 ¼ SðsÞc þ F0ðsÞ (3.20) where F0(s) represents the contribution of the surface states and the interface index S(s) ¼ dFB0/dc gives the dependence of barrier height on the metal electronegativity. The interface between metals and nonmetals has been classified into four broad types according to the resulting interfacial atomic configuration:137 (1) The nonmetal is an insulator (or a semiconductor) and the metal is physisorbed on its surface. (2) The nonmetal is a highly polarizable semiconductor and the metal makes a weak chemical bond but does not react with it to form a bulk compound. (3) The highly polarizable semiconductor reacts with the metal and forms one or more chemical compounds. (4) A thin film of oxide is left during the surface preparation of a highly polarizable semiconductor which prevents an intimate contact between the metal and the conductor. The film is referred to as interfacial layer. The type (1) interface is an ideal Schottky barrier contact, in which the barrier height varies directly with the metal work function in accordance with Eq. (3.18). The type (2) interface approximates to a Bardeen barrier,131 provided that the surface states are assumed to be distributed in space inside the semiconductor to allow a potential drop across this region. In the clean contacts of this type, one would expect the barrier height to show a weak dependence on Fm. The type (3) interface represents a case of strong chemical bonding between the metal and the semiconductor and, hence, we would expect the barrier height to depend on some quantity related to chemical or metallurgical reactions at the interface. The type (4) contact is the one which is most frequently encountered in actual metal- semiconductor devises. In most metal-semiconductor contacts, the semiconductor surface before metal deposition is prepared by chemical cleaning and a thin insulating oxide 102 Sung Pil Lee and Chowdhury Shaestagir

t Insulating interfacial film

eΦB0 Ec EF EF eΦ 0

Eν

Metal Semiconductor Figure 3.16 Electron energy band diagram of a metal-semiconductor contact with surface states and interfacial layer. layer is invariably left on the surface of the semiconductor. The thickness of this interfacial layer depends on the method of surface preparation and, for a good Schottky contact, must be less than about 20 Å. The energy band diagram of a contact with interfacial oxide layer is shown in Fig. 3.16.In this ﬁgure, potential drops linearly across the interfacial oxide layer because this layer is assumed to be an ideal insulator devoid of any charge. It has also been assumed that in the lower edge of the conduction band, the insulator lies below the vacuum level. When the interfacial layer is thin enough (<30 Å), the potential drop across it is negligibly small compared to that in the semiconductor depletion region. Such a thin layer is transparent to the electrons as the electrons can tunnel through it in either direction. For these reasons, the barrier height FB0 and the contact potential difference Vbi remain almost unaffected by the presence of a thin interfacial layer. In this state, the charge in the depletion region vanishes and the charge on the metal side is balanced by the charge in the interface states on the semiconductor side. This ﬂat band barrier height is given by138 * ¼ ð cÞ þ ð Þð Þ ¼ þ FB0 C1 Fm 1 C1 Eg F0 C1Fm C2 (3.21) where ε ¼ i C1 2 εi þ e tDs Here, εi ¼ εrε0 is the permittivity of the insulating layer, t is its thickness, e is the electronic charge, and Ds is the density of interface states per unit area The effect of electrode-oxide interfaces in gas sensor operation 103

per eV. The position of the neutral level F0 is measured from the top of the valence band. From Eq. (3.21) it is seen that as Ds tends to zero, C1 becomes * unity and FB0 tends to the Schottky limit of Eq. (3.18). The expression for the flat band barrier height is modified when there is a charge present in the interfacial oxide layer. Considering the case of the n-type semiconductor, if there is a fixed charge Qox per unit area within the oxide layer, then Eq. (3.21) is replaced by

* ¼ ð cÞ þ ð Þð Þ C1tQox FB0 C1 Fm 1 C1 Eg F0 (3.22) εi

3.3.3 Image force effects on the barrier height The other effect that makes the barrier height depend on the electric field in the depletion region is the lowering of the image force barrier. This is not dependent on the presence of the interfacial oxide layer and occurs even when such a layer is absent. Lowering of the image force barrier can be understood by referring to Fig. 3.17. When an electron is at a distance x from the metal, there exists an electric field perpendicular to the metal surface. This field may be calculated by assuming a hypothetical positive image charge e located at a distance (ex) inside the metal. The force of 2 2 attraction F between the electron and its image charge is e /4pεd(2x) , 2 and the electron has a negative potential energy F$x ¼ ee /16pεdx relative to that of an electron at infinity, as shown by the dotted curve in Fig. 3.17. This potential energy must be added to the barrier energy eeEx to obtain the total energy of the electron. It can be seen from Fig. 3.17 that the

(a) (b) Metal Dielectric E(x)

xm x

ΔΦ eε eΦB0

eΦBm EF x x x = 0 = 0 Figure 3.17 (a) Image charge and electric ﬁeld lines at a metal-dielectric interface and (b) electron energy diagram showing the image force lowering of the barrier. 104 Sung Pil Lee and Chowdhury Shaestagir

maximum energy occurs at a distance xm from the metal surface, and it can also be shown that the magnitude DFB0 (¼eDfB0) of the barrier lowering is given by26 " # 1=4 e3N DF ¼ d ðV V Þ (3.23) B0 p2ε2ε bi 8 d s

Here, Nd is the donor concentration in the semiconductor and V is the applied voltage. The image force permittivity εd may be different from the static permittivity εs of the semiconductor. This is because the electron approaches the barrier with the thermal velocity (z107 cm/s) and, if its transit time through the barrier region is small compared to the dielectric relaxation time, then the semiconductor does not get fully polarized. However, it is found that in practical situations, the electron transit time through the barrier region is sufﬁciently large to justify εd ¼ εs. The image force lowering of the barrier results from the ﬁeld produced by an electron and will be absent when there is no electron present in the semiconductor conduction band near the top of the barrier. Hence, when the barrier height is measured by a method which does not require movement of the electron over the barrier, the obtained value of FB0 is not lowered by the image force.

3.4 Charge carrier transport in the electrode-oxide semiconductor interfaces 3.4.1 Electric ﬁeld and capacitance in the metal- semiconductor interface The electric ﬁeld and potential distribution in the depletion region of a Schottky barrier junction depend on the barrier height, the applied voltage, and the impurity concentration. These dependences are frequently needed and can be obtained by the solution of a one-dimensional Poisson equation. Fig. 3.18(a) shows the energy band diagram of a reverse-biased Schottky barrier junction made on an n-type semiconductor. We assume the semiconductor to be nondegenerated and uniformly doped and divide it into a space charge region and a neutral region devoid of any space charge. At any point in the semiconductor, the Poisson equation can be written as124,125 2f d ¼e ½ þ ð Þ ð Þ 2 Nd p x n x (3.24) dx εs The effect of electrode-oxide interfaces in gas sensor operation 105

Metal Semiconductor

eΦ eVd B0 Ec E F EF W

Ev (a)

x E (b)

x φ(x) (c)

Figure 3.18 Electric field and potential distribution in the metal-semiconductor interface: (a) energy band diagram, (b) electric field, and (c) potential distribution. where εs is the semiconductor permittivity, Nd is the donor concentration, and n(x) and p(x) are the electron and the hole concentrations at any point x in the semiconductor, respectively. It is assumed that all the donors are ionized. Taking the potential f to be zero in the neutral bulk region of the semiconductor at the edge of the space charge layer, one can write efðxÞ efðxÞ nðxÞ¼n exp ; pðxÞ¼p exp (3.25) 0 kT 0 kT where n0 and p0 represent the equilibrium electron and hole concentrations in the neutral semiconductor. Substituting the values of n(x) and p(x)inEq. (3.24), one obtains 2f fð Þ fð Þ d ¼e e x þ e x 2 Nd n0 exp p0 exp (3.26) dx εs kT kT A closed form solution of this equation is not possible. An additional simplifying assumption made in the analysis is the so-called “depletion approximation.” In this approximation, the free carrier concentrations are assumed to fall abruptly from their equilibrium values n0 and p0 in the bulk neutral region to a negligibly small value in the barrier space charge 106 Sung Pil Lee and Chowdhury Shaestagir region. In reality, this transition occurs smoothly over a distance in which the bands bend by about 3kT, but the calculations made using the depletion approximation are sufficiently accurate for most purposes. Thus, using the depletion approximation, Eq. (3.26) can be written as 2f d ¼e < < 2 Nd 0 x W dx εs (3.27) ¼ 0 x > W where W represents the width of the depletion region. Integrating Eq. (3.27) with respect to x and using the condition that df/dx ¼ 0at x ¼ W,we obtain the electric field ε(x) in the depletion region: d4 x εðxÞ¼ ¼ ε 1 (3.28) dx m w where ε ¼eNd W is the maximum electric field which occurs at x ¼ 0. A m εs second integration with the boundary condition f ¼ 0atx ¼ W leads to the following relation: eN x 2 fðxÞ¼ dW 2 1 (3.29) 2εs W Thus, the potential varies parabolically with the distance in the depletion region and has its maximum value f(0) ¼ Vd given by

eNd 2 Vd ¼ðVbi V Þ¼ W (3.30) 2εs where V is the externally applied voltage. For a forward bias V ¼ VF and for a reverse bias V ¼ eVR. The negative sign in the above equation shows that the potential at x ¼ 0 is negative with respect to that at x ¼ W. The depletion region width W is obtained from the above relation: 1=2 2εs W ¼ jVbi V j (3.31) eNd The width of depletion region at zero bias is obtained by putting V ¼ 0. From Eq. (3.31), it can be seen that W decreases below its value W0 in case of a forward bias and increases above W0 in the event of a reverse bias. Figs. 3.18(b) and (c) show the electric field and the potential distributions for the Schottky barrier junction. In the depletion approximation, we have neglected the electron and hole concentrations in comparison to the donor concentration Nd. In a strong The effect of electrode-oxide interfaces in gas sensor operation 107 n-type semiconductor, the hole concentration is negligible but at the edge of the depletion region x ¼ W the electron concentration n(W) ¼ n0 ¼ Nd and decreases exponentially with the decreasing potential f(x). It should be clear that the concentration n(x) becomes negligibly small compared with Nd when ef(x)inEq. (3.25) is e4kT or less. Thus, the depletion approximation is valid only when the potential drop Vd across the depletion region is large compared to about 4kT/e. A change in the voltage across the Schottky barrier junction causes a change in the width of the depletion region, and this change is accomplished by the movement of charge carriers into the space charge layer or out of this region. This change in the depletion region charge gives rise to a capacitance. Referring to Fig. 3.18(a) and ignoring the charge in the surface states, there are three sources of charge in the barrier region. First, there is the charge Qd in the depletion region, which results from the movement of electrons out of the semiconductor into the metal. Second, there is a charge Qm on the metal surface, which is caused by the electrons that have crossed from the semiconductor into the metal. Finally, if the band bending is sufficiently large, a charge Qh will occur due to holes which exist in the semiconductor region immediately adjacent to the metal contact.95 Electrical neutrality in the junction region requires that Qd þ Qm þ Qh ¼ 0, where each of these charges represents charge per unit area of the junction. Suppose now that the bias across the junction is increased by a small amount DVd. The space charge layer capacitance C0 per unit area is defined by the relation:

0 dQd d C ¼ ¼ ðQm þ QhÞ (3.32) dVd dVd Let us ignore the effect of minority carriers and take Qh ¼ 0 so that Qd ¼ eQm. Applying Gauss law at the metal-semiconductor boundary, we obtain

εsEm ¼ Qd (3.33) The maximum electric ﬁeld strength is assumed to occur at x ¼ 0. The ﬁeld Em has been calculated in Eq. (3.28), assuming the depletion approximation. A more accurate expression for εm is obtained by integrating Eq. (3.26), assuming that band bending is small so that p(x) is negligible every- where and 2f fð Þ d ¼e e x 2 Nd n0 exp (3.34) dx εs kT 108 Sung Pil Lee and Chowdhury Shaestagir

Multiplying both sides of this relation by 2df/dx and integrating from x ¼ 0tox ¼ W, with f(0) ¼ eVd and f(W) ¼ 0 and assuming Nd ¼ n0, we obtain f 2 d ¼ ε2 ¼ 2e kT m ε Nd Vd (3.35) dx x¼0 s e where Vd ¼ (Vi V) is the voltage drop across the depletion region and the built-in voltage Vi is taken to be positive. The depletion region charge Qd per unit area is then given by = kT 1 2 Q ¼ ε ε ¼ 2eε N V (3.36) d s m s d d e and the depletion region capacitance C can be written as = dQ eε N 1 2 C ¼ A d ¼ A s d (3.37) dVd 2½Vbi ðkT=eÞV where A represents the area of the Schottky barrier contact. As charge Qd varies with the voltage in a nonlinear manner, the capacitance is a nonlinear function of voltage and can be deﬁned only as a differential capacitance presented to a small change DVd in the voltage of the depletion region. If an interfacial oxide layer is present between the metal and the semiconductor, a portion of the applied voltage appears across this layer and modiﬁes the dependence of the depletion region charge Qd on the applied voltage. The capacitances of the interfacial layer and the depletion region are effectively in series, and the overall capacitance in general may be a complicated function of the interfacial layer parameters and the applied voltage. However, when the interfacial layer is thin (about 30 Å or less), electrons can tunnel through the layer from the metal to the semiconductor and the resulting Schottky barrier is nearly ideal. The currentevoltage characteristics of the nearly ideal diode are described in terms of an ideality factor n given by 1 vF ¼ B (3.38) n evV where vFB/vV represents the change in the barrier height with applied voltage because of the presence of the interfacial layer. If the states at the oxide-semiconductor interface are uniformly distributed in energy, then n is nearly independent of the applied voltage. The effect of electrode-oxide interfaces in gas sensor operation 109

To obtain the CeV characteristics for the nearly ideal case, we substitute eVi ¼ (FB Fn)inEq. (3.36) and write

1=2 Qd ¼½2εiεsNdðFB Fn kT eV Þ (3.39) and 1=2 dQd εsNd vFB C ¼ A ¼ eA 1 (3.40) dV 2½FB0 Fn kT eV evV which, after substituting for vFB0/vV in terms of the ideality factor n, becomes 1=2 dQd eA εsNd C ¼ A ¼ (3.41) dV n 2½FB Fn kT eV

The barrier height FB for a constant n can be written as vF 1 F ¼ F þ B V ¼ F þ 1 eV (3.42) B B0 vV B0 n where FB0 represents the zero bias barrier height. Combining Eqs. (3.41) and (3.42), one obtains

1 2nA2 ¼ ½ ðF F Þ 2 2 n B0 n kT eV (3.43) C e εsNd From Eq. (3.43), one observes that the effect of interfacial layer is to scale up both the slope and the intercept. Thus, the 1/C2 versus voltage V plot underestimates the dopant concentration and results in a larger intercept V0 over its values in ideal case without interfacial layer. For a nearly ideal Schottky barrier, n is 1.1 or less so that the value of Nd is not changed much but V0 is signiﬁcantly increased. 3.4.2 Transport mechanism across the junction barrier Fig. 3.19 schematically depicts these processes for a forward biased Schottky barrier made on an n-type semiconductor. The inverse processes occur under reverse bias. The current ﬂows in a Schottky barrier because of charge transport from the semiconductor to the metal, or vice versa. There are four different mechanisms by which the carrier transport can occur: (1) thermionic emission over the barrier; (2) tunneling through the barrier; 110 Sung Pil Lee and Chowdhury Shaestagir

Δeφ B (a) (b) eVd E eφ x c B m EF

eVF EF (c) (d)

Figure 3.19 Energy band diagram of a forward biased Schottky barrier junction on n-type semiconductor showing different transport processes. (3) carrier recombination (or generation) in the depletion region; and (4) carrier recombination in the neutral region of the semiconductor, which is equivalent to the minority carrier in junctions.125 Referring to Fig. 3.19, an electron emitted over the barrier from semiconductor into the metal must move through the high field depletion region. In traversing this region, the motion of the electron is governed by the drift and the diffusion processes. The emission of electrons into the metal is controlled by the density of available states in the metal. Thus, the two processesdthe emission over the barrier and the drift and diffusion in the depletion regiondare effectively in series and whichever offers the higher resistance determines the current. In their original treatment, Wanger,139 Schottky, and Spenke140 assumed that the current was limited by the drift and diffusion processes. The diffusion theory leads to the following expression for the diode current:34 F eV I ¼ eAN mE exp B exp 1 (3.44) c m kT kT where A is the diode cross-sectional area, Nc is the effective density of states in the conduction band of the semiconductor, m is the electron mobility, and all other symbols have their usual meanings. As the maximum field Em in Eq. (3.44) is voltage-dependent, the preexponential factor in this equation does not saturate as it should in an ideal Schottky diode. Subsequent work by Bethe showed that the diode current is limited by thermionic emission over the barrier and is not in agreement with Eq. (3.44). The difference between the two mechanisms is shown by the position of the quasi-Fermi level in the depletion region. According to the diffusion theory, the electrons are in equilibrium with the lattice even when the junction is forward biased, so The effect of electrode-oxide interfaces in gas sensor operation 111 that their quasi-Fermi level coincides with the metal Fermi level at the interface, as shown by dotted curve in Fig. 3.19. In the thermionic emission theory, on the other hand, the electrons entering the metal have energy higher than the metal electrons and their quasi-Fermi level is almost horizontal through the depletion region, as shown by the dashed curve. The effect of drift and diffusion in the depletion region is assumed to be negligible in the thermionic emission theory and the barrier height is assumed to be large compared to kT. From Fig. 3.19, it is obvious that only those electrons whose kinetic energy exceeds the height of the potential barrier will be able to reach the top of the barrier. Assuming that the electrons have a Maxwellian distribution of velocities, the number of electrons n* per unit area, which have sufficient energy to move over the barrier from the semiconductor into the metal, is given by eðV V Þ n * ¼ n exp bi (3.45) 0 kT where n0 represents the electron concentration in the neutral semiconductor outside the depletion region and V is the voltage applied to the semiconductor. For a nondegenerate semiconductor, F n ¼ N exp n (3.46) 0 c kT and as FB ¼ eVi þ Fn from Eq. (3.45) we obtain * ðF eV Þ n ¼ N exp B (3.47) c kT If these electrons are assumed to have an isotropic distribution of velocities, then from the kinetic theory the flux of electrons incident on the barrier is n*v 4. Supposing that all the incident electrons cross over into the metal and none is reflected back, the current ISM due to passing of electrons from the semiconductor to the metal is given by eAv F eV I ¼ N exp B (3.48) SM 4 c kT where v is the average thermal velocity of electrons in the semiconductor. Although the electrons flow from the semiconductor into the metal, the current ISM flows from the metal to the semiconductor and is taken to be positive in Fig. 3.19. 112 Sung Pil Lee and Chowdhury Shaestagir

For unbiased junction under thermal equilibrium, no net current can ﬂow. Consequently, the current given by Eq. (3.48) must be balanced by an opposite current IMS due to crossing of electrons from the metal into the semiconductor making I ¼ ISM þ IMS ¼ 0 and eAv F I ¼ N exp B (3.49) MS 4 c kT In the presence of an applied bias V, the barrier for electron ﬂow from the metal to semiconductor remains practically unchanged at fB and so is the current IMS ¼ eI0. The current ISM, however, is given by Eq. (3.48) and, combining this equation with Eq. (3.49), we obtain eV I ¼ I0 exp 1 (3.50) kT 1=2 For a Maxwellian distribution, the average velocity v ¼ 8kT pm* 2 3/2 and substituting Nc ¼ 2(2pm*kT/h ) , the current I0 can be written as F I ¼ ART 2 exp B (3.51) 0 kT where 4pm*ek2 R ¼ h3 is the Richardson constant for thermionic emission from the metal into the semiconductor with electron effective mass m*, h is the Plank’s constant, and A is the diode area.

3.4.3 Tunneling effects in the oxide-semiconductor interface The thermionic emission diffusion theory of Crowell and Sze141 assumes that the electron distribution function remains Maxwellian in the barrier depletion region and that the classical drift and diffusion equations can be used throughout this region. Near the top of the barrier, the electric field is very high and the distribution function changes considerably within a mean free path. Under these conditions, it is no longer possible to split the current into drift and diffusion components. Moreover, we should not expect the distribution function near the top of the barrier to be Maxwellian and isotropic. As the electrons entering the metal are assumed not to return, The effect of electrode-oxide interfaces in gas sensor operation 113 the distribution function near the boundary must be anisotropic. However, the transport equations derived from the first-order solution of the Boltz- mann equation are found to be inadequate near the top of the barrier and a new set of transport equations has been proposed. These considerations show that the problem of hot electrons in a rapidly varying field has not been solved to date and we do not yet have an exact theory of the Schottky barrier currentevoltage characteristics. It is, therefore, surprising that in spite of the drastic assumptions the existing theory has been so successful in describing the IeV characteristics. Besides the diffusion and thermionic emission mechanisms, electrons can also be transported across the barrier by quantum mechanical tunneling. The two ways in which tunneling can occur in a Schottky barrier junction are shown in Fig. 3.20 for (1) forward bias and (2) reverse bias. The semiconductor in Fig. 3.20 is assumed to be doped to degeneracy such that the Fermi

(a)

TFE

e φ B Em

FE EF φn eVF Ec

(b)

Δe φ B

TFE

e φ B

EF FE eVR EF

Ec Figure 3.20 Field emission and thermionic field emission tunneling through a Schottky barrier on n-type semiconductor: (a) forward bias and (b) reverse bias. 114 Sung Pil Lee and Chowdhury Shaestagir level lies above the bottom of the conduction band. Because of heavy doping, the depletion region is very thin and, at low temperatures, electrons with energy close to the Fermi level can tunnel from the semiconductor into the metal. This process is known as “field emission” (FE). At higher temperatures, a significant number of electrons are able to rise high above the Fermi level, where they see a thinner and lower barrier. These electrons thus can tunnel into the metal before reaching the top of the barrier. This tunneling of thermally excited electrons is known as “thermionic field emission” (TFE). As the number of electrons decreases rapidly with energy above the Fermi level, and the barrier thickness and height also decreases, there exists an energy Em at which the contribution of TFE becomes maximum. If temperature is gradually raised still further, a limit is reached at which practically all the electrons are able to reach the top of the barrier and thermionic emission predominates. Tunneling through a Schottky barrier has been analyzed theoretically by Padovani and Stratton142 and by Crowell and Rideout.143 The main results of their study are described below. Field emission in the forward direction occurs only in degenerate semiconductors and, except for very low forward biases, the IeV characteristic in the presence of tunneling can be described by the relation eV I ¼ Is exp (3.52) E0 where E E ¼ E cot h 00 0 00 kT and 1=2 ¼ eh Nd E00 * 4p m εs where m* is the electron effective mass and h is Planck’s constant. The preexponential factor Is in Eq. (3.52) is only weakly dependent on voltage and is a complicated function of barrier height, parameters of the semiconductor, and the temperature. The ratio kT/E00 is a measure of the relative importance of TE and tunneling. At low temperature, Eoo may become large compared to kT and we have E0 z E00, then the slope of lnI versus the V plot is constant and independent of T. This is the case for FE. At high temperatures, where E00 kT,wegetE0 ¼ kT and the slope of the The effect of electrode-oxide interfaces in gas sensor operation 115 lnI versus the V plot is e/kT, which corresponds to TE. For intermediate values of temperature, the slope can be written as e/nkT with E E n ¼ 00 coth 00 (3.53) kT kT

3.4.4 Structure of the interfacial layer A barrier of metal/semiconductor contact is a limiting situation which can be described as two infinite half planes of material, a metal and a semiconductor, brought into contact. A typical technical barrier would result from contacting metal on the semiconductor after a series of in-ambient preparations. An ideal barrier results from depositing the metal in a carefully controlled way, where precautions are taken to keep the interface atomically clean, that is, the only atoms present are those of the initial semiconductor and the desired metal. The procedure for cleaning the surface may strongly influence the outcome by altering the surface structure stoichiometry or by introducing surface defects. Two regimes are (a) Building up the metal layer by layer by evaporation deposition (b) Pressing two bulk pieces together to form a point contact In both cases, the technical interface would have some amount of extra- neous material trapped at the interface. In actual situations, this material might consist of a native oxide of 10e15 Å thick.26 The interplay of a variety of effects then determines the interfacial composition, interfacial width, and the number of interfacial states. The complexity of the general problem arises from the diversity of concurrent effects that can be at work in specific material situations. Schottky barriers are fabricated with the layer-by-layer approach or the deposition of the metal on the semiconductor. Before treating the details of the barrier evolution stages, it is worth describing the details of the overall process to motivate the division into the stages, enumerated in Table 3.4, into which the evolution can be decomposed.144 The details of the formation of the interfacial layer are the focus of this section without concurrently emphasizing the processes that generate interface stated. This separation better allows the individual aspects to be identified, but should not imply a decoupling of the two phenomena. The interface layer can have a multiplicity of zones with a division between metallic and semiconducting regions, as depicted schematically in e Fig. 3.21.137,138,145 148 On the metal side, the alloyed zone may be of sufficient width to become the metal forming the barrier. There is an insulating layer present, 116 Sung Pil Lee and Chowdhury Shaestagir

Table 3.4 Enumeration of the layer-by-layer stages of Schottky barrier formation.

0. Clean surface a. Ordered b. Disordered 1. Dilute limit a. Chemisorption b. Reaction 2. Metal nucleation a. Eutectic formation b. Compound formation c. Interdiffusion and interaction 3. Asymptotic overlayer

Interfacial zones

Bulk 1 1 2 Bulk

Metal Semiconductor Figure 3.21 Schematic depiction of metal-semiconductor interfacial zones. either intentionally or unintentionally, between the metal and the semiconductor. This layer can be a purposefully deposited layer or a thermally grown layer. The layer could also inadvertently result from the act of junction formation. There can be an interfacial or transition layer, extending into the semiconductor due to out-diffusion of constituents of the semiconductor, in-diffusion of metal atoms, chemical interactions, and semiconductor surface damage. The presence of insulating layer can have two effects on barrier formation: one is a geometrical effect (it can simply further separate the charge in the metal from the charge in the semiconductor, giving a large dipole length) and the other is a potentially significant modification of the dipole arrangement. If the interfacial layer contains no charge, then its effect on barrier formation would be geometrical. However, the presence of an The effect of electrode-oxide interfaces in gas sensor operation 117 interfacial layer together with interface states has more than a geometrical effect. It can strongly modify the dipole arrangement because these localized states can hold charge. Such interface states can be extrinsic arising from defects caused by cross diffusion, chemical interaction, and semiconductor surface damage and rearrangement. Interface states can also be intrinsic: they can be a basic feature of the semiconductor surface or they can arise from the extension of metal electron wave function into the semiconductor gap. As interface states can store charge, they can modify the field in the interface layer. It is seen that their presence, together with the interface layer, can modify Vb and FB. Unlike the simple geometrical case, it is now possible to increase Vb and FB or decrease them, depending on the charge stored in the interface states.145 On the semiconductor side, the interface states are distributed within the two zones, one being derived from the evanescent tail of the metal wave function extending into the semiconductor and the remaining interface layer being the second zone. The interface states in zone 1 in many cases have the most significant effect on determining the barrier height. The first zone is at most 10 Å thick from estimates available in literature.145,146 The second zone can be very large depending on the thermal history of the sample and can also significantly affect the applied bias performance of the barriers. This new interfacial metal can have different characteristics from the originally deposited metal.

3.4.4.1 State 0: the clean semiconductor surface The layer-by-layer evolution starts from the clean semiconductor surface. The clear semiconductor surfaces can be divided into two types of regime: ordered and disordered. In the case of cleaved surfaces, disordering usually results from structured damage induced by the cleaving process employed. The noncleavage faces are typically cleaned by a sputtering technique which usually leaves the surface disordered and, in the case of compound semiconductors, nonstoichiometric. Thermal annealing will usually restore the ordering. The most perfect noncleavage surfaces are those grown in situ with MBE techniques. Most vacuum-semiconductor interfaces reconstruct or relax so that they show lateral periodicities or coordinates different from a strict truncation of the bulk lattice. A variety of driving forces for these reconstructions have been identiﬁed, but in general they can be summed up as deriving from the surface lowering its energy by remaining semiconducting. The 118 Sung Pil Lee and Chowdhury Shaestagir reconstruction has a strong inﬂuence on the spectrum of intrinsic surface states and their energy position. The character of surface states observed with electron spectroscopy is very sensitive to the presence of adsorbates or overlayers.

3.4.4.2 Stage 1: the dilute limit The initial stage of interface development constitutes the dilute limit. The dilute limit is characterized by a density of metal atoms small enough that a continuous metal film does not form. The structural aspects of the dilute limit are complex and depend strongly on the semiconductor during deposition. Changes in the surface structure at this stage of deposition are known as “impurity stabilized.” Deposition rate effects are not well- explored but have been demonstrated with respect to nucleation of single crystal metals. An important aspect of the dilute limit is that the heat of condensation of the metal atom may be sufficient to dislodge semiconductor surface atoms. If there is a tendency to surface reaction, then this also strongly influences this stage.

3.4.4.3 Stage 2: monolayer formation The monolayer formation stage plays the most important part in determining the characteristics of the interface. As the density of atoms increases, metal nuclei begin to form. The nucleation of the metal ﬁlm is a very complex subject. Characteristic of the monolayer formation stage is the production of interdiffusion due to the heat released as the metal nucleates. This heat augments electromigration due to the heat released as the presence of the dipole. In some cases, the release of this heat can drive other reactions over their activation barrier. The approach to monolayer formation may have several substages at which ordered metal overlayers may result. The structural aspects of the monolayer stage are strongly dependent on the deposition temperature and subsequent annealing history.

3.4.4.4 Stage 3: addition monolayers and interdiffusion The final characteristics of the interface can continue to be driven, as additional monolayers are added. Strain fields due to lattice mismatch and grain boundary phenomena can play an important role. Usually the interface reaches a stable configuration with 3e10 monolayers for most metals on semiconductor. Interfaces are often metastable, however, and the interdiffusion is accelerated at increased temperature or under biased conditions. The effect of electrode-oxide interfaces in gas sensor operation 119

The application of core level photoemission to this problem conﬁrmed what had already been deduced with Auger spectroscopy, but on a ﬁner dimensional scale. An overriding fact of metal-semiconductor interfaces is the intrinsic aspects of interfacial interdiffusion and interaction. These two aspects of the chemistry often tend to overwhelm issues that might be deduced from the clean separate surfaces. The interdiffusion can result in interface layers with radically different characteristics from the supposed components.

3.5 Gas/solid interactions in the electrode-oxide semiconductor interfaces 3.5.1 Dipole formation in the interfacial layer It is known that a number of metalsde.g., Pd or Ptdadsorb and dissolve hydrogen and that the adsorbed atoms change the work function of the metal surfaces. By using a thin metal ﬁlm as the catalytic metal electrode of the Schottky devices, hydrogen and other gases that react with the catalytic metal can be detected. Hydrogen molecules are adsorbed and dissociated on the outer metal surface. Then, the atomic hydrogen permeates through the bulk lattice toward the metal-semiconductor, causing a perturbation at this interface that gives rise to a change in the sensor output e signal.41,42,149 152 Two hypothetical steps were proposed to explain the detection mechanism:42 (1) the hydrogen atom is polarized at the metal-semiconductor interface, which gives rise to a dipole layer, or (2) an excess of charge states at the metal-semiconductor interface is created in the presence of hydrogen and reduces the Schottky barrier height. Fig. 3.22 shows the band diagrams without and with hydrogen adsorption. The hydrogen-polarized layer, arising from the intrinsic electric ﬁeld of the diode, redistributes the charges in the depletion region and abates the

(a) (b)

eΦB – + eΦB Ec – + Ec E E F EF F EF

Eν Eν

Metal Semiconductor Metal Semiconductor Figure 3.22 Energy band diagram of a Schottky junction without (a) and with hydrogen adsorption (b). 120 Sung Pil Lee and Chowdhury Shaestagir degree of band bending. Hence, the corresponding current is correlated with the number of hydrogen atoms adsorbed at the metal-semiconductor interface. When atomic hydrogen has been formed on the outer metal surface, which is exposed to the ambient, an equilibrium is reached between the hydrogen concentration at this metal surface and that at the metal- semiconductor interface. Based on this model, the change in the IeV characteristics and the decrease in the Schottky barrier height are strongly related to the hydrogen concentration. 3.5.2 Effects of hydrogen adsorption in the Schottky barrier junction According to the dipole model, originally developed by Lundstrom,61,96 the hydrogen sensitivity of catalytic gate devices is based on the dissociation of molecules on the catalytic metal surface and the diffusion through the metal film to form a polarized layer at the metal-insulator interface. The polarized layer gives rise to a shift of capacitanceevoltage curve of a MOS capacitor or the currentevoltage curve of a Schottky diode. Catalytic gate devices also respond to hydrogen-containing molecules such as hydrocarbons, provided that the molecules are also dissociated on the catalytic metal surface. Oxygen atoms are also dissociated on the catalytic metal surface. Water formation with oxygen atoms from oxygen-containing molecules consumes hydrogen and, therefore, decreases the sensor response. In other words, catalytic metal gates have a direct response to hydrogen and hydrocarbons, as well as an indirect response to oxygen molecules, whose effect is to decrease the direct response. The hydrogen response of field effect devices in normal air depends on several significant steps. First, the hydrogen molecule has to dissociate on the catalytic metal surface. This dissociation occurs in competition with adsorption of other molecules in the ambientdin particular, oxygen molecules. As the oxygen concentration is normally around 20%, a certain minimum hydrogen concentration is needed to give a significant hydrogen concentration on the catalytic metal surface. For nonporous Pd, this concentration is typically around 1 ppm if a couple of mV response is needed, while it is higher for nonporous Pt. When atomic hydrogen has formed on the metal surface, an equilibrium between the hydrogen concentration at the metal surface and the hydrogen concentration at the metaleinsulator interface is reached by diffusion of hydrogen atoms through the metal film. A very important part of this process is the surface reactions occurring at the The effect of electrode-oxide interfaces in gas sensor operation 121

H2 H2 O2 H2O HHHHHHHOOO

ΔV

SiO2

Figure 3.23 Surface reactions, diffusion, and trapping on the catalytic metal surface of a Pd-MIS device. metaleair interface. Hydrogen is supplied by the dissociation of hydrogen molecules from the gas phase, but there is also a back reaction due to the surface reactions between adsorbed hydrogen atoms and adsorbed oxygen atoms, resulting in water molecules that desorb rapidly from the surface at normal operation temperatures. In addition to this back reaction, adsorbed hydrogen atoms can also associate and desorb again as hydrogen molecules. These reactions are summarized in Fig. 3.23.153 When hydrogen atoms are adsorbed to the surface adsorption sites (s)of the metal, reaction is expressed as151

H2; gas þ 2s42Hs (3.54) Because oxygen is present in the air ambience, it can also be adsorbed on the metal surface and react with the adsorbed hydrogen to form OHs, which can then react with Hs to form water (H2O). The corresponding reactions are expressed as

O2; gas þ 2s42Os (3.55)

Os þ Hs4OHs þ s (3.56)

Hs þ OHs4H2O þ 2s (3.57)

2Hs þ Os4H2O þ 3s (3.58) Except some of the dissociated hydrogen atoms reacting with the oxygen molecules, surface oxygen, and hydroxyl group, the other adsorbed 122 Sung Pil Lee and Chowdhury Shaestagir hydrogen atoms diffuse through the metal into the metal-oxide interface and then further penetrate into the oxide ﬁlm. þ 4 þ si Hs s Hsi (3.59) þ 4 þ sO Hs s Hso (3.60) where si and so are the hydrogen adsorption sites in the interface and the oxide layer, respectively. The adsorption sites at the interface are located at the oxygen atoms on the oxide surface,153 and the absorption sites in the 104 oxide layer are located at the silicon or oxygen atoms (SiO2 devices). Because the oxygen atoms existing at the oxide surface provide more trap sites for forming the hydrogen dipoles, a large number of hydrogen atoms could then be adsorbed at the metal-oxide, thereby acting as a dipole layer. F Consequently, the resultant barrier height B;H2 is reduced and becomes smaller than fB,air because of the formation of dipoles. The hydrogen atoms absorbed within the oxide layer react with the oxide atoms and subsequently release the electrons, which raises the Fermi level of the oxide. Then, the oxide barrier VO; air between metal and oxide is gradually reduced down to

VOB;H2 during the reaction process.

3.5.3 Adsorption of other gases in the Schottky barrier junction A semiconductor gas sensor generally consists of a thin or thick ﬁlm of particulate semiconductor material, whose macroscopic conductance is measured using a pair of electrodes with dimensions much larger than the particle size. Gas phase species can become adsorbed on the particles. If physisorption occurs, there is little direct effect on the macroscopic conductivity, although occupation of surface sites by the physisorbed species may block adsorption of other species. If chemisorptions occurs, with charge transfer between the adsorbate and the semiconductor, the charged surface layer produces bending of the conduction and valence band edges. When particles are essentially separate, with no grain intergrowth, the band bending leads to a potential barrier between neighboring particles and a change in the macroscopic conductance relative to that of a pristine ﬁlm with no adsorbate. Where some particle intergrowth has occurred, the depletion region induced by chemisorption narrows the conductive regions (necks) between grains, again leading to a decrease in macroscopic conductance, The effect of electrode-oxide interfaces in gas sensor operation 123

Surface states Gas adsorption, and chemical reactions Catalytic effects charge transfer H2O CO CO C H CO2 – 2 CO x y Gas OH O– H O OH– O– 2 e– e–

Electrode Electrode

Grain boundary

Current flow Energy

eVs

x Figure 3.24 Schematic of a conductometric gas sensor with electrodes placed high up in the sensor face. relative to the ﬁlm with no adsorption.45 In either case, it will be seen that changes in the number of species adsorbed, or in their relative charges, will lead to changes in the macroscopic conductance, as depicted in Fig. 3.24.46 It is the basis of the gas sensing mechanism in conductometric (“chemresis- tor”) sensors based on tin dioxide and other semiconductors. For the detection of CO gas, the operation of sensor requires oxygen as a constituent of the background gas. The reaction of CO with the adsorbed oxygen and/or oxygen in the oxide semiconductor causes the reactant prod- 38,154 uct, CO2, to be desorbed from the device. The desorption and reaction of CO gas at steady-state proceeds via the following reaction at the metal- oxide semiconductor interface: þ 4 e 2e O2 2Oads (3.61) þ e 4 þ CO Oads CO2;des e (3.62) Oxygen can dissociate on the metal and oxygen atoms so formed may spill over onto the oxide semiconductor surface, a well-known 124 Sung Pil Lee and Chowdhury Shaestagir phenomenon in catalysis.155,156 The adsorption (reduction) of the chemi- e sorbed O ion at the metal-oxide semiconductor interface increases (decreases) the resistance of the oxide semiconductor. In MIS gas (CO, NOx, etc.) sensors which have the metal-oxide semiconductor gate layer, the behavior and gas adsorption mechanism are somewhat different from those of conventional noble gate hydrogen sensors. The chemisorption process leads to the creation of dipole moments and changes of work function, thus directly shifting the threshold voltage due to the change of carriers in the inversion layer. The consequent change in voltage across the dipole layer can be solved using Poisson’s equation, the result of which is given by105

DVth ¼Nqd=ε (3.63) where Nq is the density of the adsorption sites, d is the dipole moment of the dipole layer, and ε is the permittivity. The change in voltage across the dipole bilayer leads to a shift of Vth in the negative gate bias direction. In the IeV characteristics of the MISFET, IDsat can be expressed as CðV DV Þ2 I ¼ G th (3.64) Dsat 2 where C is a constant determined by the device design.

3.6 Conclusions Identifying the phenomena that occur between the electrode and semiconductor of the semiconductor gas sensor has been a very interesting topic for decades. One-electrode sensors using micromachining technology are widely applied, replacing the traditional interdigit two-electrode type. As a planar design is applied to the one-electrode sensor that is produced with the micromachining technology, large-scale integration is appropriate. In addition, it has advantages of low power consumption and production cost. For the microsensor represented as the FET type, materials used for the gate electrode and the type of conﬁguration are essential. When it works simply as a device that supplies voltage to the gate, poly Si or aluminum can be used. When the gate electrode should perform a catalytic function or allow gas penetration, special materials are used or a unique design is applied. Many studies also attempt to apply conductive polymer to the gas sensor. The conductive polymer electrode using polypyrrole (PPy) and polyaniline (PANI) is manufactured with casting, layer-by-layer deposition, spin-coating, or LB techniques. The effect of electrode-oxide interfaces in gas sensor operation 125

The phenomena at the electrode-semiconductor interface, accompanied by potential barrier formation, can be explained by the charge transport theory; nevertheless, several different physical situations occur, depending on the types and shapes of sensors. It is not one single mechanism that will account for all the different material situations encountered. Although the electrode of gas sensors is known to affect sensing mechanism of semiconductor gas sensors, it is still unclear how much the electrode influences the sensors. The anticipation that the electrode will improve the properties of gas sensors lead scholars researching on gas sensors to choose an optimum electrode material and modify electrode geometry according to its sensor construction. In the near future, totally new mechanisms that cannot be explained with current model may be proposed and applied for nanosized gas sensors currently being researched. References 1. Seiyama T, Kato A, Fulishi K, Nagatani M. A new detector for gaseous components using semiconductive thin films. Anal Chem 1962;34:1502e3. 2. Taguchi N. Gas detecting device. US Patent 3695848. 1972. 3. Lopez-G andara C, Ramos FM, Cirera A. Ysz-based oxygen sensors and the use of nanomaterials, a review from classical models to current trends. J Sens 2009:1e15. 4. Moos R. A brief overview on automotive exhaust gas sensors based on electroceramics. Int J Appl Ceram Technol 2005;2:401e13. 5. Riegel J, Neumann H, Wiedenmann HM. Exhaust gas sensors for automotive emission control. Solid State Ion 2002;152e153:783e800. 6. Ivers-Tiffee E, H€ardtl KH, Menesklou W, Riegel J. Principles of solid state oxygen sensors for lean combustion gas control. Electrochim Acta 2001;47:807e14. 7. Anderson T, Ren F, Pearton S, Kang BS, Wang HT, Chang CY, Lin J. Advances in hydrogen, carbon dioxide, and hydrocarbon gas sensor technology using GaN and ZnO-based devices. Sensors 2009;9:4669e94. 8. Chang YC, Bai H, Li SN, Kuo CN. Bromocresol green/mesoporous silica adsorbent for ammonia gas sensing via an optical sensing instrument. Sensors 2011;11:4060e72. 9. Wang Y, Tong MM, Zhang D, Gao Z. Improving the performance of catalytic combustion type methane gas sensors using nanostructure elements doped with rare earth cocatalysts. Sensors 2011;11:19e31. 10. Wongchoosuk C, Wisitsoraat A, Phokharatkul D, Tuantranont A, Kerdcharoen T. Multi-walled carbon nanotube-doped tungsten oxide thin films for hydrogen gas sensing. Sensors 2010;10:7705e15. 11. Fine GF, Cavanagh LM, Afonja A, Binions R. Metal oxide semi-conductor gas sensors in environmental monitoring. Sensors 2010;10:5469e502. 12. Lee SP. Micro-humidity sensor system using CMOS technology. Sens Lett 2011;9: 1e6. 13. Song K, Wang Q, Liu Q, Zhang H, Cheng Y. A wireless electronic nose system using aFe2O3 gas sensing array and least squares support vector regression. Sensors 2011;11: 485e505. 14. Lilienthal AJ, Loutfi A, Duckett T. Airborne chemical sensing with mobile robots. Sensors 2006;6:1616e78. 126 Sung Pil Lee and Chowdhury Shaestagir

15. Gonzalez-Jimenez J, Monroy JG, Blanco JL. The multi-chamber electronic nosedan improved olfaction sensor for mobile robotics. Sensors 2011;11:6145e64. 16. Endres HE, Gottler€ W, Hartinger R, Drost S, Hellmich W, Muller€ G, Braunmuhl€ CB, Krenkow A, Perego C, Sberveglieri G. A thin-film SnO2 sensor system for simultaneous detection of CO and NO2 with neural signal evaluation. Sens Actuators B 1996;36:353e7. 17. Lee SP. Synthesis and characteristics of carbon nitride for micro humidity sensors. Sensors 2008;8:1508e18. 18. Chen D, Lei S, Chen Y. A single polyaniline nanofiber field effect transistor and its gas sensing mechanisms. Sensors 2011;11:6509e16. 19. Schuler€ M, Sauerwald T, Schutze€ A. Metal oxide semiconductor gas sensor self-test using Fourier-based impedance spectroscopy. J Sens Syst 2014;3:213e21. 20. Kang KS, Lee SP. CO gas sensors operating at room temperature. J Mater Sci 2003;38: 4319e23. 21. Hagleitner C, Lange D, Hierlemann A, Brand O, Baltes H. CMOS single-chip gas detection system comprising capacitive, calorimetric and mass-sensitive microsensors. IEEE J Solid State Circuits 2002;37:1867e78. 22. Liu J, Wang W, Li S, Liu M, He S. Advances in SAW gas sensors based on the condensate-adsorption effect. Sensors 2011;11:11871e84. 23. Marr I, Reiß S, Hagen G, Moos R. Planar zeolite film-based potentiometric gas sensors manufactured by a combined thick-film and electroplating technique. Sensors 2011;11:7736e48. 24. Chinvongamorn C, Pinwattana K, Praphairaksit N, Imato T, Chailapakul O. Amper- ometric determination of sulfite by gas diffusion-sequential injection with boron- doped diamond electrode. Sensors 2008;8:1846e57. 25. Pascoe KJ. Properties of materials for electrical engineers. London, UK: John Wiley & Sons; 1973. 26. Rhoderick EH. Metal-semiconductor contacts. Oxford, UK: Clarendon Press; 1978. 27. Sze SM. Physics of semiconductor devices. New York (NY), USA: John Wiley & Sons; 1981. 28. Sze S. Semiconductor devices. New York (NY), USA: John Wiley & Sons; 2002. 29. Zhou X, Xu Y, Cao Q, Niu S. ‘Metalesemiconductor ohmic contact of SnO2-based ceramic gas sensors. Sens Actuators B Chem 1997;41:163e7. 30. Schierbaum KD. Engineering of oxide surfaces and metal/oxide interfaces for chemical sensors, recent trends. Sens Actuators B Chem 1995;24e25:239e47. 31. Zhang W, Vascincelos EA, Uchida H, Katsube T, Nakatsubo T, Nishioka Y. A study of silicon Schottky diode structures for NOx gas detection. Sens Actuators B 2000;65: 154e6. 32. Irokawa Y. Hydrogen sensors using nitride-based semiconductor diodes, the role of metal/semiconductor interfaces. Sensors 2011;11:674e95. 33. Schalwig J, Muller€ G, Karrer U, Eickhoff M, Ambacher O, Stutzmann M, Gorgens€ L, Dollinger G. Hydrogen response mechanism of PteGaN Schottky diodes. Appl Phys Lett 2002;80:1222e4. 34. Petty MC. Conduction mechanisms in Pd/SiO2/n-Si Schottky diode hydrogen detectors. Solid State Electron 1986;29:89e97. 35. Matsuo K, Negoro N, Kotani J, Hashizume T, Hasegawa H. Pt Schottky diode gas sensors formed on GaN and AlGaN/GaN heterostructure. Appl Surf Sci 2005;244: 273e6. 36. Soo MT, Cheong KY, Noor AFM. Advances of SiC-based MOS capacitor hydrogen sensors for harsh environment applications. Sens Actuators B 2010;151:39e55. 37. Morrison SR. Selectivity in semiconductor gas sensors. Sens Actuators B 1987;12: 425e40. The effect of electrode-oxide interfaces in gas sensor operation 127

38. Morrison SR. Mechanism of semiconductor gas sensor operation. Sens Actuators B 1987;11:283e7. 39. Barsan N, Huebner M, Weimar U. Conduction mechanism in semiconducting metal oxide sensing film, impact on transition. In: Jaaniso R, Tan OK, editors. Semiconductor gas sensors. Cambridge, UK: Woodhead Publishing; 2013. p. 64e113. 40. Liu X, Cheng S, Liu H, Hu S, Zhang D, Ning H. Review a survey on gas sensing technology. Sensors 2012;12:9635e65. 41. Potje-Kamloth K. Semiconductor junction gas sensors. Chem Rev 2008;108:367e99. 42. Hubert T, Boon-Brett L, Black G, Banach U. Hydrogen sensorsda review. Sens Ac- tuators B 2011;157:329e52. 43. Korotcenkov G. ‘Practical aspects in design of one-electrode semiconductor gas sensors’, Status report. Sens Actuators B 2007;121:664e78. 44. Kasap S. Principles of electrical engineering materials and devices. New York: McGraw-Hill; 1997. 45. Hummel R. Electronic properties of materials. New York: Springer-Verlag; 1985. 46. Capone S, Siciliano P, Quaranta F, Rella R, Epifani M, Vasanelli L. Moisture influence of geometry effect of Au and Pt electrodes on CO sensing response of SnO2 microsensors based on sol-gel thin film. Sens Actuators B 2001;77:503e11. 47. Gourari H, Lumbreras M, Van Landschoot R, Schoonman J. Electrode nature effects on stannic oxide type layers prepared by electrostatic spray deposition. Sens Actuators B 1999;58:365e9. 48. Saukko S, Latto V. Influence of electrode material on properties of SnO2-based gas sensor. Thin Solid Films 2003;436:137e40. 49. Durrani SM. The influence of electrode metals and its configuration on the response of tin oxide thin film CO sensor. Talanta 2006;68:1732e5. 50. Pijolat C. Etude des proprietes physico-chimiques et des proprietes electriques du dioxyde d’etain en fonction de l’atmosphere gazeuse environnante, Application a la detection selective des gaz [Ph.D. thesis]. Polytechnique de Grenoble; 1986. 51. Bertrand J, Koziej D, Barsan N, Viricelle JP, Pijolat C, Weimar U. Influence of the nature of the electrode on the sensing performance of SnO2 sensors, Impedance spectroscopy studies. In: Proceedings of eurosensors XX, Goeteborg, Sweden, 17e20 September 2006; 2006. p. 100e1. 52. Ylinampa A, Lantto V, Lepp€avuori S. Some differences between Au and Pt electrodes in SnO2 thick-film gas sensors. Sens Actuators B 1993;13e14:602e4. 53. Rank S, Hafner S, Barsan N, Weimar U. The impact of the nature of the electrode material on SnO2 thick Film sensor performance, Influence on oxygen adsorption. Process Eng 2012;47:514e7. 54. Barsan N, Schweizer-Berberich M, Gopel€ W. Fundamental and practical aspects in the design of nanoscaled SnO2 gas sensors, a status report. J Anal Chem 1999;365: 287e304. 55. Lin HM, Tzeng SJ, Hsiau PJ, Tsai WI. Electrode effects on gas sensing properties of nanocrystalline zinc oxide. Nanostruct Mater 1998;10:465e77. 56. Bender F, Kim C, Mlsna T, Vetelino JF. Characterization of a WO3 thin film chlorine sensor. Sens Actuators B 2001;77:281e6. 57. Sofian M, Oussama ME, Imad AA, Marsha CK. Semiconducting metal oxide based sensors for selective gas pollutant detection. Sensors 2009;9:8158e96. 58. Ahmad IA. Metal/metal-oxide nanoclusters for gas sensor applications. J Nanomater 2016;16:1e717. 59. Steele M, MacIver B. Palladium/cadmium-sulfide Schottky diodes for hydrogen detection. Appl Phys Lett 1976;28:687e1687. 60. Toohey MJ. Electrodes for nano dot-based gas sensors. Sens Actuators B 2005;105: 232e50. 128 Sung Pil Lee and Chowdhury Shaestagir

61. Lundstrom I, Shivaraman MS, Svensson S. A hydrogen sensitive Pd-gate MOS transistors. J Appl Phys 1975;46:3876e81. 62. Ito K. Hydrogen sensitive Schottky barrier diodes. Surf Sci 1979;86:345e52. 63. Cheng C, Tsai Y, Lin K, Chen H, Hsu W, Chuang H, Chen C, Liu W. Hydrogen sensing characteristics of Pd- and Pt-Al0.3Ga0.7As metal semiconductor(MS) Schottky diodes. Semicond Sci Technol 2004;19:778e82. 64. Kim J, Gila B, Chung G, Abernathy C, Pearton S, Ren F. Hydrogen sensitive GaN Schottky diodes. Solid State Electron 2003;47:1069e73. 65. Basu S, Roy S, Jacob C. Ruthenium as Schottky metal for SiC-based high temperature hydrogen sensors. Mater Technol Hydrog Econ 2004;801:193e8. 66. Salehi A, Nazerian V. Characterization of magnetic Ni/n-Si Schottky contact for hydrogen gas sensing applications. Sens Actuators B 2007;122:572e7. 67. Pandis C, Brilis N, Bourithis E, Tsamakis D, Ali H, Krishnamoorthy S, Iliadis A, Kompitsas M. Low-temperature hydrogen sensors based on Au nanoclusters and Schottky contacts on ZnO films deposited by pulsed laser deposition on Si and SiO2 substrates. IEEE Sens J 2007;7:448e54. 68. Hossein-Babaei F, Purahmad M. Wide dynamic range hydrogen sensing using silver- rutile Schottky diode. In: Proceedings of the third international conference on sensing technology, Taipei, Taiwan, 30 Novembere03 December 2008; 2008. p. 72e5. 69. Song J, Lu W, Flynn J, Brandes G. AlGaN/GaN Schottky diode hydrogen sensor performance at high temperatures with different catalytic metals. Solid State Electron 2005;49:1330e4. 70. Golovanov V, Liu CC, Kiv A, Fuk D, Ivanovskaya M. Microfabricated one-electrode In2O3 and Fe2O3 composite sensors. Solid State Phys 2009;13:68e73. 71. Tamaki J, Miyaji A, Makinodan J, Ogura S, Konishi S. Effect of micro-gap electrode on detection of dilute NO2 using WO3 thin film microsensors’. Sens Actuators B 2005; 108:202e6. 72. Shaalan NM, Yamazaki T, Kikuta T. Influence of morphology and structure geometry on NO2 gas-sensing characteristics of SnO2 nanostructures synthesized via a thermal evaporation method. Sens Actuators B 2011;153:11e6. 73. Meixner H, Lampe U. Metal oxide sensors. Sens Actuators B 1996;33:198e202. 74. Korotcenkov G. Handbook of gas sensor materials. New York (NY), USA: Springer; 2013. 75. Hoefer U, Kuhner€ G, Schweizer W, Sulz G, Steiner K. CO and CO2 thin-film SnO2 gas sensors on Si substrates. Sens Actuators B 1994;22:115e9. 76. Michel HJ, Leiste H, Halbritter J. Structural and electrical characterization of PVD- deposited SnO2 films for gas esensor application. Sens Actuators B 1995;24e25: 568e72. 77. Sozza A, Dua C, Kerlain A, Brylinski C, Zanoni E. Long-term stability of Ti-Pt-Au metallization system for Schottky contact and first-level metallization on SiC MESFET. Microelectron Reliab 2004;44:1109e13. 78. Capone S, Epifani M, Francioso L, Kaciulis S, Mezzi A, Siciliano P, Taurino A. Influ- ence of electrodes ageing on the properties of the gas sensors based on SnO2. Sens Actuators B 2006;115:396e402. 79. Lee J, Kim E. Effect of structural and morphological changes on the conductivity of stretched PANI-DBSA/HIPS film. Bull Korean Chem Soc 2011;32:2661e5. 80. Skotheim TA. Handbook of conducting polymers. New York (NY), USA: Marcel Dekker; 1986. 81. Ray S, Easteal AJ, Cooney RP, Edmonds NR. Structure and properties of melt-processed PVDF/PMMA/polyaniline blends. Mater Chem Phys 2009;113: 829e38. The effect of electrode-oxide interfaces in gas sensor operation 129

82. Singh SK, Gupta RK, Singh RA. Optical, mechanical, and electrical studies of polymer composites based on charge transfer complex of phenothiazine-iodine with poly(vinyl chloride). Synth Met 2009;159:2478e85. 83. Zhang DH. On the conductivity measurement of polyaniline pellets. Polym Test 2007; 26:9e13. 84. Ansari R, Raofie F. Removal of mercuric ion from aqueous solutions using sawdust coated by polyaniline. Eur J Chem 2006;3:35e43. 85. Holmes P, Loasby R. Handbook of thick film technology. Glasgow, UK: Electrochemical Pub.; 1976. 86. Song K, Joo B, Choi N, Lee Y, Lee S, Huh J, Lee D. A micro hot-wire sensors for gas sensing applications. Sens Actuators B 2004;102:1e6. 87. Korotcenkov G, Boris I, Brinzari V, Lychkovsky Y, Karkotsky G, Golovanov V, Cornet A, Rossinyol E, Rodriguez J, Cirera A. Gas sensing characteristics of one- electrode gas sensors on the base of doped In2O3ceramics. Sens Actuators B 2004; 103:13e22. 88. Jones E. The pellistor catalytic gas sensor. In: Moseley P, Tofield B, editors. Solid-state gas sensors. Bristol, UK: Institute of Physics; 1987. p. 17e31. 89. Fukui K. Detection and measurements of odour by sintered tin oxide. Sens Actuators B 1991;5:27e32. 90. Malchenko S, Lychkovski Y, Baykov M. In2O3-based gas sensors. Sens Actuators B 1993;13e14:159e61. 91. Malchenko S, Lychkovski Y, Baykov M. One electrode semiconductor sensors for detection of toxic and explosive gases in air. Sens Actuators B 1992;7:505e6. 92. Faglia G, Comini E, Sberveglieri G, Rella R, Siciliano P, Vasanelli L. Square and collinear four probe array and Hall measurements on metal oxide thin film gas sensors. Sens Actuators B 1998;53:69e753. 93. Williams D. Semiconducting oxides as gas-sensitive resistors. Sens Actuators B 1999;57: 1e16. 94. Gerlich M, Kornely S, Fleischer M, Meixner H, Kassing R. Selectivity enhancement of a WO3/TiO2 gas sensor by the use of a four-point electrode structure. Sens Actuators B 2003;93:503e8. 95. Bl€aser D, Ruhl€ T, Diehl C, Ulrich M, Kohl D. Nanostructured semiconductor gas sensors to overcome sensitivity limitations due to percolation effects. Phys A 1999; 266:218e23. 96. Lundstrom I. Hydrogen sensitive MOS structures, Part 1, Principles and applications. Sens Actuators B 1981;1:403e26. 97. Abe H, Esashi M, Matsuo T. ISFET’s using inorganic gate thin films’. IEEE Trans Electron Devices 1979;26:1932e9. ED. 98. Sibbald A, Whalley P, Covington A. A miniature flow-through cell with a four- junction chemFET integrated circuit. Anal Chem Acta 1984;159:47e62. 99. Bergveld P. Development of an ion-sensitive solid state device for neuro-physiological measurements. IEEE Trans Biomed Eng 1990:70e1. BME-17. 100. Mariucci L, Fortunato G, Pecora A, Bearzotti A, Carelli P, Leone R. Hydrogenated amorphous silicon technology for chemical sensing thin film transistors. Sens Actuators B 1992;6:29e33. 101. Litovchenko VG, Gorbanyuk TI, Efremov AA, Evtukh AA. Effect of macrostructure and composition of the top metal electrode on properties of MIS gas sensors. Microelec- tron Reliab 2000;40:821e4. 102. Qiu YY, Azeredo-Leme C, Alcacer LR, Franca JE. A CMOS humidity sensor with on-chip calibration. Sens Actuators A 2001;92:80e7. 103. Gu L, Huang Q, Qin M. A novel capacitive-type humidity sensor using CMOS fabrication technology. Sens Actuators B 2004;99:491e8. 130 Sung Pil Lee and Chowdhury Shaestagir

104. Eriksson M, Salomonsson A, Lundstrom€ I. The influence of the insulator surface properties on the hydrogen response of field-effect gas sensors. J Appl Phys 2005;98. 0349031e0439036. 105. Fukuda H, Kasama K, Nomura S. Highly sensitive MISFET sensors with porous Pte SnO2 gate electrode for CO gas sensing applications. Sens Actuators B 2000;64:163e8. 106. Trinchi A, Kandasamy S, Wlodarski W. High temperature field effect hydrogen and hydrocarbon gas sensors based on SiC MOS devices. Sens Actuators B 2008;133: 705e16. 107. Johnson CL, Wise KD, Schwank JW. A thin-film gas detector for semiconductor process gases. IEDM Tech Dig 1988:662e3. 108. Suehl JS, Cavicchi RE, Gaitan M, Semancik S. Tin oxide gas sensor fabricated using CMOS micro-hotplates and in situ processing. IEEE Electron Device Lett 1993;4: 118e20. 109. Chan PCH, Yan G, Sheng LY, Sharma RK, Tang Z, Sin JKO, Hsing IM, Wang YY. An integrated gas sensor technology using surface micro-machining. In: Proceedings of the MEMS 2001, Interlaken, Switzerland, 21e25 January, 2001; 2001. p. 543e6. 110. Yan G, Tang Z, Chan PCH, Sin JKO, Hsing IM, Wang Y. An experimental study on high-temperature metallization for micro-hotplate-based integrated gas sensors. Sens Actuators B 2002;86:1e11. 111. Dai CL, Xiao FY, Juang YZ, Chiu CF. An approach to fabricating microstructures that incorporate circuits using a post-CMOS process. J Micromech Microeng 2005;15: 98e103. 112. Dai CL. A capacitive humidity sensor integrated with micro heater and ring oscillator circuit fabricated CMOS-MEMS technique. Sens Actuators B 2007;122:375e80. 113. Janata J, Josowicz M. Suspended gate field effect transistors modified with polypyrrole as alcohol sensor. Anal Chem 1986;58:514e7. 114. Hauptman P. Sensors, principles and applications. Hertfordshire, UK: Prentice Hall; 1991. 115. Lee SP, Park KJ. Humidity sensitive field effect transistors. Sens Actuators B 1996; 35e36:80e4. 116. Lee KN, Seo YT, Yoon S, Lee MH, Kim YK, Seong WK. Chemical gating experiment of a nano-field-effect transistor sensor using the detection of negative ions in air. Sens Actuators B 2016;236:654e8. 117. Jain U, Harker A, Stoneham A, Williams D. Effect of electrode geometry on sensor response. Sens Actuators B 1990;2:111e4. 118. Williams D, Pratt K. Theory of self-diagnostic sensor array devices using gas-sensitive resistors. J Chem Soc Faraday Trans 1995;91:1961e6. 119. Vilanova X, Llobet E, Brezmes J, Calderer J, Correig X. Numerical simulation of the electrode geometry and position effects on semiconductor gas sensor response. Sens Ac- tuators B 1998;48:425e31. 120. Gardner J. Intelligent gas sensing using an integrated sensor pair. Sens Actuators B 1995; 26e27:261e6. 121. Hoefer U, Steiner K, Wagner E. Contact and sheet resistances of SnO2 thin films from transmission-line model measurements. Sens Actuators B 1995;26:59e63. 122. Hoefer U, Bottner€ H, Felske A, Kuhner€ G, Steiner K, Sulz G. Thin-film SnO2 sensor arrays controlled by variation of contact potentialeeA suitable tool for chemometric gas mixture analysis in the TLV range. Sens Actuators B 1997;44:429e33. 123. Reeves G, Harrison H. Obtaining the specific contact resistance from transmission line model measurements. IEEE Electron Device Lett 1982;3:111e3. 124. Neamen D. Semiconductor physics and devices. New York (NY), USA: McGraw-Hill; 2003. The effect of electrode-oxide interfaces in gas sensor operation 131

125. Tyagi M. Physics of Schottky barrier junction. In: Sharma B, editor. Metal- semiconductor Schottky barrier junction and their application, NY, USA, Plenum, New York; 1984. 126. Schottky W. Naturwissenchaften 1938;26:843. 127. Mott N. Note on the contact between a metal and an insulator or semiconductor. Proc Camb Philol Soc 1938;34:568e72. 128. Islam M, Hakim M. Electron affinity and work function of polycrystalline SnO2 thin film. J Mater Sci Lett 1986;5:63e5. 129. Feucht D. Heterojunctions in photovoltaic devices. J Vac Sci Technol 1977;14:57e64. 130. Gutmann S, Wolak M, Conrad M, Beerbom M, Schlaf R. Effect of UV and X-ray radiation on the work function of TiO2 surfaces. J Appl Phys 2010;107:103705. 131. Halas S, Durakiewicz T. Work functions of elements expressed in terms of the Fermi energy and the density of free electrons. J Phys Condens Matter 1999;10:10815e25. 132. Methfessel M, Hennig D, Scheffler M. Trends of the surface relaxations, surface energies, and work functions of the 4d transition metals. Phys Rev 1992;B13:4816e29. 133. Akbi M. A method for measuring the photoelectric work function of contact materials versus temperature. IEEE Trans Compon Packag Manuf Technol 2014;4:1293e302. 134. Kal W, Frese Jr . Simple method for estimating energy levels of solids. J Vac Sci Technol 1979;16:1042e4. 135. Bardeen J. Surface states and rectification at a metal-semiconductor contact. Phys Rev 1947;71:717e27. 136. Milnes A, Feucht. Heterojunctions and metal-semiconductor junctions (chap. 6 and chap. 7). New York: Academy Press; 1972. 137. Mead C. Metal-semiconductor surface barriers. Solid State Electron 1966;9:1023e32. 138. Andrew J, Phillips J. Chemical bonding and structure of metal-semiconductor interfaces. Phys Rev Lett 1969;22:1433e6. 139. Wagner C. Theory of current rectifier. Phys Z 1931;32:641e5. 140. Schottky W, Spenke E. Quantitative treatment of the space charge and boundary- layer theory of the crystal rectifier. Wiss Veroff Siemens-Werken 1939;18:225e91. 141. Crowell C, Sze S. Current transport in metal-semiconductor barrier. Solid State Electron 1966;9:1035e48. 142. Padovani F, Stratton R. Field and thermionic-field emission in Schottky barriers. Solid State Electron 1966;9:695e707. 143. Crowell C, Rideout V. Normalized thermionic-field emission in metal- semiconductor barriers. Solid State Electron 1969;12:89e105. 144. Bachrach RZ. Interface chemistry and structure of Schottky barrier formation. In: Sharma B, editor. Metal-semiconductor Schottky barrier junction and their application. New York, USA: Plenum; 1984. 145. Heine V. Theory of surface states. Phys Rev 1965;A138:1689e96. 146. Louie SG, Chelikowsky JR, Cohen ML. Ionicity and the theory of Schottky barriers. Phys Rev B 1977;15:2154e62. 147. Cowley A, Sze S. Surface states and barrier height of metal-semiconductor systems. J Appl Phys 1965;36:3212e20. 148. Wada O, Majerfeld A, Robson P. InP Schottky contacts with increased barrier height. Solid State Electron 1982;25:381e7. 149. Lundstrom I, Svensson C. In: Janata J, Huber J, editors. In solid state chemical sensors. New York: Academic Press; 1985. 150. Lee C, Yan J. Sensing mechanisms of Pt/b-Ga2O3/GaN hydrogen sensor diodes. Sens Actuators B 2010;147:723e9. 151. Ruths P, Ashok S, Fonash S, Ruths J. A study of Pd-Si MIS Schottky-barrier diode hydrogen detector. IEEE Trans Electron Devices 1981;ED-28:1003e9. 132 Sung Pil Lee and Chowdhury Shaestagir

152. Johansson M, Lundstrom I, Ekedahl L. Bridging the pressure gap for palladium metal- insulator-semiconductor hydrogen sensors in oxygen containing environments. J Appl Phys 1998;84:44e51. 153. Lundstrom€ I, Sundgren H, Winquist F, Eriksson M, Krantz-Rulcker C, Lloyd- Spetz A. Twenty-ﬁve years of ﬁeld effect gas sensor research in Linkoping. Sens Actu- ators B 2011;157:329e52. 154. Eggins B. Chemical sensors and biosensors. New York: Wiley; 2002. p. 49. 155. Yanagida H. Gas sensing characteristics of porous zinc oxide ceramic with and without platinum catalyst. In: Proc. int. meet. on chemical sensors; 1983. p. 95e100. 156. Pajonk G, Teichner S, Germain J. Studies in surface science and catalysis. New York: Elsevier; 1983. p. 17. CHAPTER FOUR

Introduction to semiconductor gas sensors: a block scheme description

Arnaldo D’Amico, Corrado Di Natale Department of Electronic Engineering, University of Rome Tor Vergata, Roma, Italy

Contents

4.1 Introduction 133 4.2 The sensor blocks 135 4.2.1 The response curve 136 4.2.2 Sensitivity 137 4.2.3 Resolution 138 4.2.4 Example of the evaluation of resolution 139 4.3 Metal oxide semiconductor capacitor: the case of the hydrogen gas sensitivity of 142

Pd-SiO2-Si 4.4 Light-addressable potentiometric sensor 144 4.5 Metal oxide semiconductor ﬁeld-effect transistor 148 4.6 Metal oxide semiconductors 151

4.6.1 SnO2 bands 151 4.6.2 Band diagram modulation 153 4.7 Conclusions 156 References 156

4.1 Introduction Several human activities require the measure of the concentration of gases and volatile compounds. Industrial processes and pollution control have been the traditional applications of gas sensors.1 In the last years, a number of novel and fascinating applications emerged, not least the detection of the volatile metabolites as biomarker for several diseases.2 In this context, analytical instruments are ready to provide the necessary accuracy for monitoring gases and vapors. However, although the great efforts of miniaturization of traditional instruments, such as gas chromatographs and

Semiconductor Gas Sensors, Second Edition ISBN: 978-0-08-102559-8 © 2020 Elsevier Ltd. https://doi.org/10.1016/B978-0-08-102559-8.00004-5 All rights reserved. 133 j 134 Arnaldo D’Amico and Corrado Di Natale mass spectrometers,3,4 these are still bulky, expensive, and require trained and skill technicians for proper operation and for their data analysis and interpretation. Solid-state sensors have been always considered the solution to overcome some of these drawbacks and to facilitate gas analysis avoiding the mandatory use of dedicated laboratories. Furthermore, solid-state sensors can take advantage of the micro/nanoe- lectronic technologies; thus, they can be cheap and reliable and furthermore readily available for the latest information technologies such as the so-called Internet of things.5 Semiconductor gas sensors are solid-state devices based on semiconductor materials and/or semiconductor devices. Then, when we talk of semiconductors, we may intend either those materials (inorganic or organic) that show some intrinsic sensitivity to gases or the devices that constitute the basis for transducers such as, for instance, those based on field effects where changes in the electronic properties of the gate can control either the output current or voltage. The parameters of devices and materials depend on the quality and quantity of the gases and vapors to which they are in contact. Semiconductor materials, particularly in transducer systems, offer the perspective of integration of the sensors in the microelectronic realm taking advantage of the integrated electronic interfaces for signal processing. In this paper we will consider a variety of sensors. Before describing the working principles of sensors, it is worthy to introduce some basic definitions of sensor properties.6,7 These definitions define a common language necessary for an objective comparison among sensors. Ideal gas sensors should be sensitive, selective, and stable. Additional properties are reversibility, accuracy, response time, and linearity. For many applications, additional properties such as low cost, small size, unsensitivity to radiation and temperature, and robustness are also required. Finally, sensors should always be compatible with standard preprocessing electronics. In practice, being almost impossible to satisfy at once all these constraints an acceptable compromise is always sought satisfying only a part of the above-mentioned features. In this paper, we will describe the sensors through a block scheme which helps to decompose the sensor principle in a number of elementary steps each endowed with its own sensitivity. The scheme might help to compare different sensors emphasizing where the different performance arises. Thus, instead of paying attention to the basic theories, widely available in literature Introduction to semiconductor gas sensors: a block scheme description 135 and in many chapters of this book, this chapter stays focused on the basic intrinsic sensitivities. To facilitate the comprehension of the basic mechanisms of the sensors based on field effects, we will consider as an example the chemical sensitivity of palladium films with respect to the hydrogen gas.8 This example points out the basic properties of the devices and it can be easily generalized to semiconductor devices made of either inorganic or organic materials.

4.2 The sensor blocks Fig. 4.1 shows a complete block scheme of a general sensor system. Not only the whole system can be considered as a sensor but also elements or subsets of the scheme can be identiﬁed as sensors itself. The whole appraisal of such a scheme is mandatory to understand the role of devices. Sensors are the interface between the electronic circuits and the outer world. It is simple to understand that the properties of sensor signals, namely the electric signals depending on the outer world quantities, are determined both by the features of the sensor as a device and by the electric circuit at which the sensor is connected.

v,i Outer world sensor Y electronic v,i v,i A/D amplifier filter measurand Y=f(M) interface conversion

actuation μP energy

storage communication

display

Figure 4.1 General scheme of a sensor system. The intrinsic sensor is the interface between the electronic system and the outer world. 136 Arnaldo D’Amico and Corrado Di Natale

Fig. 4.1 shows the general scheme of a sensor-based electronic system. The intrinsic sensor element provides the interface with the outer world. The parameter Y is the electric parameter affected by the interaction with the gas. This may be, for instance, the resistivity, dielectric constant, or work function. The quantity Y itself is not observable until it is transformed by the interface circuit into a voltage or a current signal. The signal can then be amplified and filtered to facilitate its measurement. The measurement is usually performed through an analog to digital conversion, and the digital signal is processed to actuate some action on the outer world (e.g., a regulation system), stored, communicated, or displayed. Finally, it is important to keep in mind that each part of the system needs energy. All steps of the processing chain contribute to the overall sensitivity. Each block is characterized by a proper input/output relationship and the combination of all of them makes the total sensitivity. 4.2.1 The response curve The response is any quantity useful to represent the state of a sensor as a device or of the system at which the sensor is connected or of a sensor system that can be simple (formed by only one block) or rather complex made by more than one block as described in Fig. 4.1. In some cases, the response is defined in reference to a baseline value that is the response in absence of stimulus. In such a case, the response can be given as the difference, ratio, or relative change with respect to the baseline. Independently from the definition, the response is a function of the concentration of the gas. The functional relationship depends on the nature of the intrinsic sensing element and in case of a sensor system on the circuit parameters. The response curve (see Fig. 4.2) is usually represented in a Cartesian plane identified by two axes carrying the sensor signal and the gas concentration (generally called measurand), respectively. Because of the unavoidable noise, the origin of the plot cannot be reached, but rather the curve stops in correspondence of the noise level of the output signal. The response curve is the result of the sensor calibration. This operation corresponds in measuring the sensor signal when the device is exposed to known concentrations. The fitted experimental points give rise to the analytical response curve that is normally used to estimate the concentration once the output of the sensor is measured. Introduction to semiconductor gas sensors: a block scheme description 137

Vout

Vout, max

Vout, min=Vnoise M measurand min Mmax Figure 4.2 Example of a typical response curve, when the sensor response is the output voltage signal (Vout).

4.2.2 Sensitivity It is defined as the derivative of the response curve with respect to the measurand: dV S ¼ out (4.1) dM The sensitivity is the slope of the response curve (Fig. 4.2), and because the response is limited, the increase of sensitivity narrows the range of concentration. In case of gas sensors, a large sensitivity implies the capability to measure tiny changes of concentration. Where the response curve is flat, the sensitivity is null; this means that in this region any small change of the measurand does not change the output signal. Fig. 4.3 shows a subset of blocks of Fig. 4.1; it represents the steps leading from the concentration to the measured digital quantity. Taking into consideration these steps, it is convenient to express the partial sensitivities according to the nomenclature given in the same figure. It appears clearly that the overall sensitivity (the sensitivity of all the chain) is only partially related to the sensitivity of the first block, which is the most important. Rather the whole sensitivity includes also many steps of the electronic processing of the signal. The sensitivity of the intermediate blocks may be called the internal sensitivities. As shown along this chapter, the internal sensitivity can be further divided in many others, depending on the complexity of the sensor system. 138 Arnaldo D’Amico and Corrado Di Natale

dN dN dv dv dv dY df df df df df S = = × f × a × 0 × = AD × F × A × C × S dM dv f dva dv0 dY dM dv f dva dv0 dY dM Figure 4.3 The total sensitivity is given by the product of the sensitivities (derivatives of the transfer functions) of all the blocks of the chain leading from the measurand to the measured quantity.

Fig. 4.3 shows elements of a typical sensor system. The intrinsic sensor is characterized by a parameter that is affected by the measured M, the sensor is connected to an interface circuit and then any variation of Y produces a variation of a signal v0, the signal can be amplified giving rise to a larger signal va, noisy components could be removed by a filter and vf is the filtered signal, and finally the analog signal can be converted in a number (N) by an analog to digital converter. The definition of sensitivity can be applied to any block of the chain. Note that in case of the amplifier, the sensitivity is an alternate definition of the gain and the two quantities are coincident when the output of the amplifier is linearly proportional to the input. The total sensitivity (dN/dM) is then obtained with as the product of the sensitivity of each block. 4.2.3 Resolution The resolution is the necessary consequence of the peculiar nature of the quantity represented in the vertical axis of the response curve. Indeed, the sensor signal is the result of a measurement and it is subject to measurement errors. Errors of measurement are contributed by both the finite accuracy of the measurement instrument (in modern electronic systems this is ruled by the analog to digital conversion) and by the electronic noise. The electronic noise fixes the lower limit for the measurement error. The resolution (Mres) is the smallest measurable change of the measurand. Given a sensor signal Vout, the corresponding resolution is V V M ¼ lim out ¼ noise (4.2) res / Vout Vnoise S S Introduction to semiconductor gas sensors: a block scheme description 139

The sensitivity is calculated in the neighbor of the measurand at which Vout is measured. Of course, in case of a nonlinear response curve, both the resolution and the sensitivity are functions of the measurand. Vout can be considered at the output of each block in Fig. 4.3. As each electronic block adds its own noise, shorter the block chain, better the resolution. It is worth pointing out that noise and sensitivity measurements are of paramount importance for the sensor characterization to evaluate the resolution of a given sensor. Different types of noise may be encountered when we are dealing with sensors, and not all sensors have the same noise, while some sensors may show more than one kind of noise. In this context, the relevant parameter for the noise characterization is the noise spectral density s( f ). The root mean square of the output signal (Vrms) is the measurable manifestation of the noise. The relationship between S( f ) and the root mean square of the output signal is 2 3 = Z 1 2 6 f2 7 Vrms ¼ 4 sðf Þdf 5 (4.3) f1

The integral is calculated in the frequency interval practically defined by the measurement time. Shorter the measurement time, wider the frequency interval and then the noise contribution to the sensor signal. The most important types of noise are listed in Table 4.1.9 Excess noises are manifested as an additional contribution to the current and then they occur only in biased materials. It is worth mentioning that in gas sensors, additional sources of noise come from the fluctuation of the concentration at the sensor surface of the compounds at which the sensor is exposed and from the fluctuation of adsorption/desorption processes. These terms contribute to the overall spectral density of sensor noise as an additional flicker-like noise.10 4.2.4 Example of the evaluation of resolution To fix the ideas about the relationship between sensitivity and resolution, let us consider the simple circuit in Fig. 4.4 where a resistance temperature detector (RTD) is connected to a simple readout circuit. The relationship between temperature and resistance can be easily generalized to any resistive gas sensor. Let us consider the sensor represented by a linearized response curve 140 Arnaldo D’Amico and Corrado Di Natale

Table 4.1 Characteristics of the most typical noises found in semiconductors. c, c0, and c00 are constants; g is a factor close to 2; a is a factor close to 1; d is a factor ranging from 2 3; b is a factor ranging from 0.8 to 3. q is the electron charge and k is the Boltzmann constant. s is the recombination time of generationerecombination

(GR) processes. Finally, u ¼ 2pf, i is current, sv is spectral density of voltage, and si is spectral density of current. Denomination Deﬁnition Spectral density

Thermal noise Manifestation of the sv ¼ 4$k$T$R thermal motion of electric charges (electrons in solids); its magnitude is proportional to the resistance. Shot noise Excess noise typical for si ¼ 2$q$i space charge regions and therefore present in all junction devices (e.g., diodes). V g Flicker noise Excess noise likely due sv ¼ c ua to a continuous distribution of traps, typical for semiconductors. ¼ 0V d Burst noise Excess noise emerging sn c ub in semiconductors likely due to impurity atoms. ¼ 00 s GR noise Excess noise due to GR sn c 1þu2s2 processes in semiconductors.

RSðTÞ¼R0$½1 þ a$ðT T0Þ (4.4)

R0 is the resistance known at the reference temperature T0 and a is the temperature coefficient. The block scheme of the sensor in Fig. 4.4 is shown in Fig. 4.5. The RTD is made of a material whose resistivity is affected by changes of temperature. The transducer circuit that generates the sensitive signal is made by the current source I0, the signal is then amplified by the noninverting amplifier, and, in case, it can be converted into a digital quantity. The overall sensitivity of Vout with respect to the temperature can be written as Introduction to semiconductor gas sensors: a block scheme description 141

Vout (T) I0 RS(T)

Figure 4.4 RS is a resistive sensor, a resistance temperature detector (RTD) in the example, connected to a noninverting ampliﬁer.

Transducer A/D R(T) Ampliﬁer N circuit conversion (R+ΔR) (R+ΔR)*I (R+ΔR)*I*A

ρ+Δρ

ΔT ρ(T)

Figure 4.5 Block diagram of the sensor systems based on the circuit of Fig. 4.4. dVout dVout dVin dRS R2 ¼ ¼ A$ST $SR ¼ 1 þ $I0$R0a (4.5) dT dVin dRS dT R1 where A is the ampliﬁcation, ST is the sensitivity of the transducer circuit, and SR is the intrinsic sensitivity of the thermistor. The resolution can be directly calculated from the deﬁnition.

Vnoise Tres ¼ (4.6) A$ST $SR Note that the magnitude of noise depends on the single block characteristics; for instance, the amplifier amplifies both the signal and the noise and then it is irrelevant in improving the resolution, rather it worsens the resolution because it adds its own noise to the total signal. This last expression makes evident that the noise of the output signal and the sensitivity of each block are necessary to evaluate the sensors performance. On the other hand, the knowledge of the intrinsic sensor (the temperature sensor in this case), although fundamental for the final performance, does 142 Arnaldo D’Amico and Corrado Di Natale not provide any knowledge about the actual performance until the electronic blocks are defined and characterized.

4.3 Metal oxide semiconductor capacitor: the case of the hydrogen gas sensitivity of Pd-SiO2-Si The first solid-state gas sensor that we take into consideration is the metal oxide semiconductor (MOS) capacitor schematically shown in Fig. 4.6. Let us consider the case when the gate metal is palladium.11 The catalytic properties of palladium to favor the dissociative adsorption of hydrogen gas and the following diffusion of atomic hydrogen are known, as well the H2 sensing properties of this device. The basic structure is formed by a stack of four regions: an ohmic contact, p-type silicon, silicon dioxide, and the thin film of palladium. In the depletion mode, the total capacitance Ct is given by the series of the oxide and the depletion layer capacitances. In a reference atmosphere where hydrogen gas is not present, the difference between the work functions of Pd and Si (p-type) is such to generate a 12 depletion layer. As a consequence of H2 adsorption, the depletion layer size changes. This results in a variation of the depletion capacitance and then in the total capacitance. The intrinsic mechanism of sensitivity is reassumed in Fig. 4.7. The global sensitivity is the derivative of the total capacitance with respect to the hydrogen concentration. It is worth pointing out that each block is characterized by a proper response. In fact the first block is supposed to be the slowest being due to chemical reactions at the palladium surface and to the diffusion process of

Figure 4.6 Schematic of Pd-OS capacitor. The capacitance is the series of the oxide capacitance (ﬁxed) and the depletion layer capacitance (variable). The depletion layer size changes as a consequence of the exposure to H2. Introduction to semiconductor gas sensors: a block scheme description 143

Δ (H2) Δ δ ΔΦ Δ VFB

Δ xd

Δ Vout ΔCtot

Figure 4.7 The exposure to hydrogen gas concentration ([H2]) results in a dissociative adsorption of H2 at the palladium surface. It produces a concentration of atomic hydrogen that quickly diffuses toward the oxide surface. The layer of atomic hydrogen forms a dipole layer (d) at the oxide surface that can be interpreted as a change of the work function difference (DF) between palladium and silicon, the consequence of which is a change of the ﬂat band voltage (VFB) that is related to the size of the depletion layer (xd) and then to the total capacitance (DCtot). Finally, a proper transducer circuit can convert the capacitance changes into a voltage change (DVout). hydrogen atoms from the palladium surface to the Pd/SiO2 interface. The electronic time constant should not be neglected; however, the overall response time is mainly controlled by the diffusion processes, which are rather slow being the diffusion rate of atomic hydrogen through palladium of the order of 5 ms/Å. From Fig. 4.7, the total sensitivity St can be expressed as dV dV dC dx dV dDF dd S ¼ out ¼ out tot d FB (4.7) d½H2 dCtot dxd dVFB dDF dd d½H2 From the above expression, all the physical terms contributing to the overall sensitivity are visible. The last term dd/d[H2] is the intrinsic contribution of palladium, while all the other terms, summarized in dCtot/dd,are related to the MOS structure, and then they depend on the oxide thickness, permittivity of oxide, semiconductor, and the doping concentration in silicon, and the area of the MOS structure. Eventually, the ﬁrst term depends on the particular circuit used to convert the capacitance changes into a voltage change. Fig. 4.8 shows an example of a circuit that can be used to measure the value of the capacitor C as a function of the applied DC voltage V0. The MOS capacitor is biased by the sum of a DC voltage (V0) and AC voltage (vi); then the total applied voltage, if R1 ¼ R2,isVS¼ V0 Vm cos(ut). It is important that Vm V0. In this way, the AC signal is a small perturbation necessary to extract a signal proportional to C, but C depends almost completely on V0. Of course vi must be as small as possible but of a 144 Arnaldo D’Amico and Corrado Di Natale

R1 V0

Figure 4.8 Example of a circuit to measure the capacitance of metal oxide semiconductor structure as a function of the applied voltage V0. sufficient level to give a measurable output with at least a signal to noise ratio not less than about 6. The output signal is given by dV V ¼ R C s ¼ R C uV sinðutÞ (4.8) out 3 tot dt 3 tot m The above circuit can be used to measure the relationship between Ctot. and V0; this is an outmost characteristic of MOS devices, and for sensing purposes, all circuit parameters can be optimized and kept fixed to measure the changes of Ctot. The scheme shown in Fig. 4.7 describes the chain of sensitivity when the MOS capacitor is not biased in inversion. In case of inversion, the depletion layer size is fixed, rather in this case, the change of the flat band voltage affects the threshold voltage at which the inversion layer is formed. This condition is exploited in the metal oxide semiconductor field-effect transistor (MOSFET) device discussed below.

4.4 Light-addressable potentiometric sensor The intrinsic photoconductivity of semiconductors is exploited in MOS structures to give rise to an interesting device called light- addressable potentiometric sensor (LAPS).13 As shown in Fig. 4.9, the output voltage across a load resistance RL is generated by an internal equivalent voltage generator which is developed Introduction to semiconductor gas sensors: a block scheme description 145

P O

+qε D ε −qε

Figure 4.9 Principle of light-addressable potentiometric sensor. Photons absorbed in silicon give rise to electronehole couples. The couples generated in the space charge region are separated by the built-in electric field. The displacement charges the metal oxide semiconductor (MOS) capacitor and results in a signal observable across a load resistor in series to the MOS capacitor. Keeping constant the photon flux, the signal is proportional to the volume of the space charge region and then it is affected by work function changes. once pulsed light of suitable time width and of specific wavelength is absorbed in the semiconductor. In a first approximation, the electrical equivalent circuit of LAPS is shown in Fig. 4.10. Vin is the internal voltage source generated across the depletion region whose magnitude is due, by the product of the depletion region impedance and the current of the electrons and holes produced by the adsorbed photons. Electrons and holes are separated by the built-in electric field located inside the space charge region; the capacitor is the series of the oxide and the depletion capacitances, while RL is the load resistance. When the device is shined by pulses of light, the output voltage corresponds to the derivative

R Vi L

Figure 4.10 Light-addressable potentiometric sensor equivalent circuit. 146 Arnaldo D’Amico and Corrado Di Natale

Light intensity

Time

Figure 4.11 Typical output voltage of a light-addressable potentiometric sensor in case of pulsed light. of the voltage pulses (and not current pulses); the time constant is equal to CRL (see Fig. 4.11). The gas sensing mechanism is illustrated by the block scheme in Fig. 4.12. The gas sensitivity is due to the same processes occurring in the MOS capacitor; the difference here is that the size of the depletion layer modulates the amount of electronehole pairs that are separated by the built-in electric ﬁeld. In practice, instead of measuring the change of the MOS capacitor, here a voltage proportional to the ﬂux of impinging photons and to the

Δ(H2) V Δδ ΔΦ ∆ FB

Δxd vout n − p i n ; i p Couples Figure 4.12 Block diagram of light-addressable potentiometric sensor. Introduction to semiconductor gas sensors: a block scheme description 147 volume of the space charge region are generated. Actually the sensor is sensitive to both light intensity and gas concentration and the device becomes selective to gas only once the intensity of light has been kept constant. dV dV dq dx dV dDF dd S ¼ out ¼ out $ d FB (4.9) d½H2 dq dxd dVFB dDF dd d½H2 The quantity dq namely the amount of photo-induced electronehole pairs is proportional to the intensity of the impinging light. Nevertheless, the sensitivity of the LAPS is proportional to the intensity of light. In other words, the sensitivity lies in the number of electronehole couples related to the adsorbing volume defined by the depletion layer width and the section area of the device. It is worth to consider that the light pulse is a probe of the extension of the space charge region. The dependence from the light introduces an additional noise because of the fluctuation of the light source; this is equivalent to a shot noise proportional to the root square of the light intensity. In terms of sensitivity improvement, the key quantity is dxd/dVFB. The doping concentration of the semiconductor can actually determine a larger variation of the depletion region with respect to changes of the flat band voltage. LAPS sensors can be easily integrated in arrays.14 Fig. 4.13 shows a pictorial example. Each element of the array could be made of a different gate material or could be probed by light of different wavelengths. The light can be addressed sequentially to each element of the matrix. The fi n-multiple repetition of the scan and thep storageffiffiffi data may allow a signi cant noise reduction evaluated by a factor of n where n is the number of light scans per each element.

Light Pt

W Pd

Figure 4.13 Pictorial view of an array of light-addressable potentiometric sensor. 148 Arnaldo D’Amico and Corrado Di Natale

4.5 Metal oxide semiconductor field-effect transistor MOSFETs are among the most reliable and versatile transducers for gas sensor applications. In electronics the MOSFET structure has been continuously modified and improved along the years and it is at the basis of both analog and digital electronics.12 More than four decades ago, the MOSFET structure was transformed into chemical sensors replacing the gate with either an electrolyte or a catalytic metal15,16 because that MOSFET has been the basis for the development of several sensors and biosensors for a plethora of different applications. Being the MOSFET an extension of the MOS capacitor, also in this case the sensitivity is activated anytime, the interaction between the gas and gate produces an amount of charges or dipoles at the gate SiO2 interface. This layer of charges or dipoles generates an additive gate voltage and then a variation of the drain source current. Fig. 4.14 shows a schematic drawing of a MOSFET. In its first implementation as gas sensor, the gate was a film of hydrogen-sensitive palladium.12 The charge control equation of the MOSFET in the quasi-linear region is approximately12

V S VD

VG Metal Metal

Oxide Metal Oxide Oxide Inversion n+ Depletion n+

p-type Si p+ Metal

VB Figure 4.14 General scheme of a metal oxide semiconductor ﬁeld-effect transistor device. Introduction to semiconductor gas sensors: a block scheme description 149

w V 2 I ¼ m $C $ ðV V Þ$V DS (4.10) DS n ox L GS T DS 2 In the saturation region, the current becomes largely independent from VDS: w I ¼ m $C $ ðV V Þ2 (4.11) DS n ox 2L GS T The above relationship holds beyond the pinch-off point and neglects the effective channel length change due to VDS. In the above formulas, VGS and VDs are the voltages applied between the gate and source and drain and source, respectively, and IDS is the current from drain to source. mn is the effective electrons mobility in the channel, Cox is the oxide capacitance, w and L are the width and the length of the channel, respectively, and VT is the threshold voltage. This is a key parameter of the device because, at first approximation, only when VGS > VT the channel is formed and the current can flow. The threshold voltage depends on all the relevant quantities of the MOS structure such as the flat band voltage and the oxide capacitance. The saturation condition is obtained from Eq. 4.10 under the condition that the current IDS does not depend on VDS, namely dIDS/dVDS ¼ 0. This condition is achieved when VDS ¼VGS VT. Eqs. (4.10) and (4.11) are plotted in Fig. 4.15. This is the so-called output characteristics of the MOSFET. The MOSFET should be polarized to fix the quiescent point in the saturation region. The block scheme of the sensing mechanism of the Pd-FET is shown in Fig. 4.16. It corresponds to the block scheme of MOS capacitor and LAPS, except that changes in the flat band voltage in this case affect the threshold voltage. Then if the biasing parameters of the MOSFET are kept constant, the IDS changes. The original Pd-FET structure was modified to extend the sensitivity to other species than hydrogen and to use more sensitive materials. A first interesting development consisted in the use of ultrathin metal films as gate. In this condition, the metal film is not homogeneous but rather it is characterized by a number of cracks that leave the oxide exposed to the ambient air.17 In this way, the analyte does not necessitate traveling through the metal but it can reach directly the oxide surface. This offered the possibility to use more catalytic metal and to expand the sensitivity to other analytes such as ammonia.18 Additionally, a cracked gate layer can also accommodate 150 Arnaldo D’Amico and Corrado Di Natale

0.05 Vgate= 1V 0.045 Vgate=1.5V Vgate= 2V 0.04 Vgate=2.5V Vgate= 3V 0.035 Vgate=3.5V Vgate= 4V 0.03 Vgate=4.5V Vgate= 5V 0.025 Vgate=5.5V

0.02 Drain current [mA] 0.015

0.01

0.005

0 012345 Drain-source voltage [V] Figure 4.15 Currentevoltage curves of the metal oxide semiconductor ﬁeld-effect transistor where the gate voltage is kept as a parameter.

Δ(H2) ς Δδ ΔΦ ∆ ΦΒ

∆ ςΤ

∆ςουτ

Δ ΙΔΣ Figure 4.16 Block diagram of Pd field-effect transistor (FET). The adsorption of hydrogen gas changes the flat band voltage of the metal oxide semiconductor (MOS) structure, and the flat band voltage affects the threshold voltage and then the currentevoltage characteristics of the device. nonconductive organic layers as chemically sensitive material. This opportunity has been exploited to develop porphyrins functionalized MOSFETs.19 The MOSFET structure was also fabricated with organic semiconductors.20 These devices are usually made as thin film transistor architecture, and the small thickness of the organic semiconductor enables the accumulation mode and the sensitive materials as the organic semiconductor.21,22 Introduction to semiconductor gas sensors: a block scheme description 151

4.6 Metal oxide semiconductors MOS are probably the more diffused gas sensors. A number of different metal oxides can be actually used; the ﬁrst of these sensors was made of ZnO23 but the most popular of these materials them became the 24 tin oxide (SnO2). In the following section, the working principles of 25 SnO2 are illustrated. Imperfect stoichiometric ratio between tin and oxygen occurs in real materials; the vacancy of oxygen atoms leaves a quota of electrons of tin atoms not engaged in covalent bonds and weakly bond to their atoms. At room temperature, this results in a light doping of the material that shows an n-type character. The surface chemistry of tin oxide is complex, and here the simplest reactions involving oxygen are described. The principle of operation of SnO2-based semiconductor gas sensor is mainly related to the change in conductivity occurring when reducing gases (such as CO, CH4) interact with chemisorbed oxygen ions. The adsorption of O2 at the surface of tin oxide results in charged species that subtract electrons from the semiconductor conductance band. As a consequence, a surface depletion layer appears and the surface conductivity is strongly reduced. The reduction of conductivity is manifested in polycrystalline materials, where the current moves from grain to grain and then it is subject to the built-in potentials associated to the surface depletion layers. The interaction with reducing gases removes the surface oxygen species, electrons are released in the conduction band and the intergrain potential barriers are also lowered, and as a consequence, the conductivity increases. The above-described process is temperature-activated and normal operation temperatures are between 150 and 600C, depending on metal oxide and gas. Surface addition of nanoparticles deposited can greatly improve the sensitivity and the selectivity.26 More recently, organic surface functionalization was also introduced.27,28

4.6.1 SnO2 bands At thermal equilibrium (no voltage applied), many situations are possible at the SnO2 surface. As an example we consider the case of a SnO2-based CO sensor for which conduction and valence band bandings are mainly because of the following charge modiﬁers: traps (due to surface defects), additives, 152 Arnaldo D’Amico and Corrado Di Natale

D T I – Δφ Δσ

Figure 4.17 Block diagrams of metal oxide semiconductor gas sensors in case of CO, or any other reducing gas, detection. Oxygen ions adsorbed at the metal oxide surface interact with airborne CO molecules, and as a consequence, the depletion layer size and the surface barrier are decreased. In a polycrystalline material, it results in a change of the total conductivity. A transducer circuit, such as that shown in Fig. 4.4, transforms the change of conductivity into a voltage signal. adsorbed oxygen and oxygen ions formation, and formation of CO2 in presence of CO. Fig. 4.17 shows the block schemes of the sensing principle. Here, we consider two main cases, taking into account that mixed situations involving not all the modifiers may be possible. Fig. 4.18 shows a typical cross section of a SnO2 sensor with two ohmic contacts, a layer of polycrystalline SnO2 of a given thickness and an electric insulator but thermal conductor substrate and a heater on the other face. Fig. 4.19 shows the band diagrams of a monodimensional sequence of adjacent grains, in thermal equilibrium. The equilibrium condition is reached through a rearrangement of electrons across the interfaces between grains. In Fig. 4.19, the charge, in the deep depletion approximation, and the electric field are shown. When a voltage is applied across the electrodes, we can suppose that the potential is distributed only across the depletion layer. This is a typical assumption when electronic devices are modeled.12 Then the voltage drop reduces the barrier height from one side of the contact between grain, and it results in a net current flowing in the material (see Fig. 4.20). Clearly, the bulk region of the grains does not participate to the gas sensitivity or to the conduction. In nanocrystalline materials, the bulk is actually eliminated and the whole material is exploited for sensing and conduction purposes.

Electric Electric Polycrystalline material contact contact

Insulating substrate

Heater Figure 4.18 Cross section of a typical polycrystalline sensor. Introduction to semiconductor gas sensors: a block scheme description 153

(a)

(b)

(c)

Figure 4.19 (a) Band diagram, (b) charge distribution, and (c) electric ﬁeld in a monodimensional arrangement of metal oxide grains. Note that the chain is terminated at both sides by a junction with the metal electrode. and F are the work functions Fm SnO2 of the metal electrode and the sensor material, respectively. fi is the Schottky barrier between SnO2 and the metal electrode, fi is the intergrain potential barrier. ND is the electrons density in the conduction band corresponding to the donor concentration and xd is the depletion layer size at the intergrain interface.

4.6.2 Band diagram modulation Each step of the sensing principles illustrated in Fig. 4.17 in the block scheme changes the band diagram of the material at the metal oxideeair interface. Fig. 4.21 shows four steps leading from the equilibrium conﬁguration of inert material to the consequences of the interaction with a reducing gas. Here again the carbon monoxide is explicitly mentioned. Fig. 4.21(a) shows the band diagram of inert material. This condition is met either in vacuum or at a low temperature. Low temperature means less than about 150C, namely when the adsorption of airborne oxygen is not favored. At high temperature and in air, the adsorption of oxygen can take place. Oxygen can be adsorbed both as atomic and molecular species. 154 Arnaldo D’Amico and Corrado Di Natale

V=0 Fermi level

E φS – VA V>0 φS

φS – VA φ V>0 S

Figure 4.20 Changes in the conduction band diagram of the junction between two adjacent grains at the equilibrium (V ¼ 0) and under bias positive and negative. VA is the portion of the total applied voltage across a single junction. Owing to the numer- osity of grains, VA is small enough to ensure the quasi-equilibrium condition. Both thermionic and tunnel currents can be simultaneously present. Both the contributions give a current inversely proportional to the exponential of the barrier height.

In Fig. 4.21(b), the case of atomic oxygen adsorption is shown. The molecular oxygen undergoes a dissociative adsorption onto the metal oxide surface and two atomic oxygens are adsorbed. The bond is provided by two electrons which are displaced from the conduction band to the oxygen atoms. As a consequence, the surface region of the semiconductor is depleted of electrons, the bands bend upward, and a surface potential (qfS)andaworkfunction change appear. Surface oxygen can further react, at the optimal temperature, with a reducing gas molecule (such as CO). The consequence is the formation of a volatile CO2 molecule and the release of an electron in the conduction 0 band. This elicits a reduction of the surface barrier (qf S) and the work function. The full picture is much more complex because of the multiple oxygen species, each adsorbed at different energy. Furthermore, the presence of additional species in air, e.g., water vapor, makes the involved chemistry more complex. However, the above description provides a sufﬁcient introduction to the main phenomena involved in the gas sensitivity of MOS. nrdcint eiodco a esr:abokshm description scheme block a sensors: gas semiconductor to Introduction

(a)In air at low tempearture (b)In air at high tempearture (c)In air and reducing gas at (d) In air and reducing gas at high temperature: step 1 high temperature: step 2

Figure 4.21 Band diagram modulation in the different steps of reducing gas detection. (a) In air at low temperature, (b) in air at high temperature, (c) in air and reducing gas at high temperature (step 1), and (d) in air and reducing gas at high temperature (step 2). 155 156 Arnaldo D’Amico and Corrado Di Natale

4.7 Conclusions In this chapter, a general introduction to the topic of semiconductor gas sensors has been provided. General phenomena related to the sensing properties of semiconductor materials and semiconductor devices have been introduced, discussing the properties of MOS (resistors) and MOS device (capacitors) and the related FET. The discussion has been maintained at a general level focusing the attention on the basic processes responsible of the gas sensitivity. For this reason, block schemes have been introduced to help the reader to localize the sensitivity sources. The main purpose of these diagrams is to introduce a decomposition of the global sensitivity into elementary phenomena to make evident where the sensitivity emerges, which are the steps to improve the sensor, and ﬁnally how to modify the sensor to extend its property. We would like to propose this approach as a general method to present sensor properties. This in particular is necessary for novel sensors where a block scheme with the partial sensitivity values enables the comparison with other similar or, sometimes, identical sensors. A last point is concerned with the overall response time which is made up by the contribution of each block and then the knowledge of the response time of each elementary element is necessary for the development of more performant sensors. References 1. Lee D. IEEE Sens J 2001;1:214e24. 2. Di Natale C, Paolesse R, Martinelli E, Capuano R. Anal Chim Acta 2014;824:1e17. 3. Lu C, Whiting J, Sacks R, Zellers E. Anal Chem 2003;75:1400e9. 4. Black W, Stocks B, Mellors J, Engen J, Ramsey J. Anal Chem 2015;87:6286e387. 5. Potyrailo R. Chem Rev 2016;116:11877e923. 6. D’Amico A, Di Natale C. IEEE Sens J 2001;1:183e90. 7. D’Amico A, Di Natale C, Sarro P. Sens Actuators B 2015;207:1060e8. 8. Conrad H, Ertl G, Latta E. Surf Sci 1974;41:435e46. 9. Van der Ziel A. Noise in solid state devices and circuits. (New York, USA): J. Wiley; 1986. 10. Falconi C, Di Natale C, Martinelli E, D’Amico A, Zampetti E, Gardner J, Van Vliet C. Sens Actuators B 2012;174:577e85. 11. Poteat T, Lalevic B. IEEE Trans El Dev 1982;29:123e9. 12. Sze S, Ng K. Physics of semiconductor devices. 3rd ed. J. Wiley; 2006. 13. Bratov A, Abramova N, Ipatov A. Anal Chim Acta 2010;678:149e59. 14. Hu N, Ha D, Wu C, Zhou J, Kirsanov D, Legin A, Wang P. Sens Actuators A 2012;187: 50e6. 15. Bergveld P. IEEE Trans Bio-Medical Eng 1970;17:70e1. 16. Lundstrom I, Shivaraman S, Svensson C, Lundkvist L. Appl Phys Lett 1975;26:55e70. 17. Spetz A, Helmerssson U, Enquist F, Armgarth M, Lundstrom€ I. Thin Solid Films 1989; 177:77e93. 18. Spetz A, Armgarth M, Lundstrom€ I. J Appl Phys 1988;64:1274e83. Introduction to semiconductor gas sensors: a block scheme description 157

19. Andersson M, Holmberg M, Lundstrom I, Lloyd-Spetz A, Martensson P, Paolesse R, Falconi C, Proietti E, Di Natale C, D’Amico A. Sens Actuators B 2001;77:567e71. 20. Guillaud G, Al Sadoun M, Maitrot M, Simon J, Bouvet M. Chem Phys Lett 1990;167: 503e6. 21. Torsi L. Dodalabapour Anal Chem 2005;382A:381A. 22. Mabeck I, Malliaras G. Anal Bioanal Chem 2006;384:343e53. 23. Seyama T, Kato A, Fujiishi K, Nagatani M. Anal Chem 1962;34. 24. Barsan N, Weimar U. J Electroceramics 2001;7:143e67. 25. Yamazoe N, Shimanoe K. Sens Actuators B 2011;158:28e34. 26. Fine G, Cavanagh L, Afonja A, Binions R. Sensors 2010;10:5469e502. 27. Sivalingam Y, Martinelli E, Catini A, Magna G, Pomarico G, Basoli F, Paolesse R, Di Natale C. J Phys Chem C 2012;116:9151e7. 28. Hijazi M, Rieu M, Stambouli V, Tournier G, Viricelle J, Pijolat C. Sens Actuators B 2018;256:440e7. This page intentionally left blank PART TWO

Materials

159j This page intentionally left blank CHAPTER FIVE

One- and two-dimensional metal oxide nanostructures for chemical sensing

E. Comini, D. Zappa Department of Information Engineering, University of Brescia, Brescia, Italy

Contents

5.1 Introduction 161 5.2 Deposition techniques 162 5.2.1 Two-dimensional nanostructures 163 5.2.2 One-dimensional nanostructures 166 5.2.2.1 Vapor phase growth methods 167 5.2.2.2 Liquid phase growth methods 168 5.2.2.3 Template-assisted methods 169 5.3 Conductometric sensor 169 5.3.1 Device integration 170 5.4 Transduction principles and related novel devices 170 5.5 Conclusion and future trends 174 References 175

5.1 Introduction Metal oxides have very different electrical properties from metals, semiconductors, to insulators and are used in many different areas such as sensors, superconductors, magnets, and lighting. In relation to chemical sensing applications, the ability of metal oxides to change their electrical conductivity with the composition of the surrounding atmosphere has been known for almost 60 years.1 In October 1968, the ﬁrst generation of commercial devices was produced on a large scale by TGS (Taguchi Gas Sensor, now Figaro Engineering Inc.) in Japan. These sensors were made of SnO2 thick ﬁlms and were used for the detection of explosive gases. Over the years, the demand for cheap, small, low power consuming but reliable solid-state chemical sensors has continued to grow. Consequently,

Semiconductor Gas Sensors, Second Edition ISBN: 978-0-08-102559-8 © 2020 Elsevier Ltd. https://doi.org/10.1016/B978-0-08-102559-8.00005-7 All rights reserved. 161 j 162 E. Comini and D. Zappa significant research efforts have been made worldwide to improve on the “3Ss” (sensitivity, selectivity, and stability), mainly with empirical approaches, but also through some basic theoretical research and spectroscopy studies. The metal oxide that has been paid the greatest attention for chemical 2 sensing is SnO2 ; however, other n-type semiconducting oxidesdsuch as 3 4 5 6 7 TiO2, In2O3, WO3, ZnO, Fe2O3, dhave been proposed and studied. On the contrary, p-type oxidesdlike CuO8 and NiO,9dare not extensively investigated yet, mainly due to the lower performances expected.10 Further- more, the use of mixed oxides in form of heterostructures, as well as the addition of noble metals or other functional materials,11 has been studied to improve not only the sensitivity but also the selectivity and the stability. One of the most important articles in metal oxide chemical sensing was published in 1991 by Yamazoe.12 It was shown that, as the crystallite size was reduced, there was a huge improvement in sensor performance. This shifted the research community focus to investigate the performance of materials with the smallest crystallite size, while ensuring that their properties remain stable during long-term, high-temperature operation, which is necessary for metal oxide chemical sensing. Another important discovery that changed the field of chemical sensing was the synthesis of single crystal one-dimensional oxide nanowires. These materials have great potential thanks to their reduced lateral dimensions and single crystal habits, both for fundamental study and for potential nanodevice applications, i.e., the third generation of metal oxide gas sensors. The first technique used to fabricate these nanowires was the simple evaporation of the desired commercial metal oxide powders at high temperatures, followed by condensation at lower temperatures on the substrates. This technology was developed back in the 1960s and is still the most widespread growth process, even if nowadays there are many other techniques that enable the synthesis of nanostructures of different shape and size.

5.2 Deposition techniques The technique of layer deposition and coating has a variety of industrial applications, such as protective layers, sensors, resistive films, and catalyzers. We will briefly present the different deposition techniques as a function of the nanostructures obtained for , two-dimensional nanostructures (thin films), , one-dimensional nanostructures (nanowires, nanorods, etc.). One- and two-dimensional metal oxide nanostructures for chemical sensing 163

5.2.1 Two-dimensional nanostructures There are several preparation methods that have been optimized and proposed during recent years. We can, for example, distinguish between physical vapor deposition (PVD) techniques, chemical vapor deposition (CVD) techniques, and techniques that do not require a vacuum. In the case of PVD techniques, the source material to be deposited is in the solid phase. An intermediate vapor phase is then formed and finally the solid phase is deposited on the substrate. The method used for the vaporization of the source material distinguishes between the different deposition techniques. For example, heat transfer is used in thermal and electron beam evaporation (EBE), while bombardment by energetic ions is used in sputtering processes. The thermal evaporation (TE) technique is the oldest PVD process. In this technique, the substrate is kept at a short distance from the source material, which is heated until it vaporizes. The vapor then condenses on the substrate.13,14 The equipment required to apply all these PVD methods works at pressures lower than ambient pressure: this is necessary to control the composition of the deposited material. To improve its purity, the mean free path of the particles must be greater than the distance between the source and the substrate. In the specific case of TE, this also allows a lower operational temperature for the vaporization. A conventional deposition system consists of a vacuum chamber, a mechanical roughing pump, a high vacuum pump, a heated crucible, and a substrate holder. In the case of TE, unintentional doping in the film can result from the high temperature of vaporization that has to be reached by the source material. Particular attention must be paid to the material composition of the crucible containing the source material and to the possible alloys that it can form with the source material at the evaporation temperature. EBE is similar to TE, except that the source material is vaporized by the heat transferred from an electron beam accelerated toward the source. On the impingement of the high-energy electrons, their kinetic energy is converted into heat and the source material can reach temperatures exceeding 3000C, causing a local melting where the beam is focused. The advantage of this technique is that the electron beam can be focused on specific areas of the source material, so that the interaction between e source material and support materials can be reduced.15 17 164 E. Comini and D. Zappa

Another method based on evaporation is pulsed laser deposition. The high photon flux incident on the target induces an essentially instantaneous e temperature increase that causes the evaporation of the target material.18 20 The experimental setup consists of a vacuum chamber in which there is the target material and a window through which the laser beam is focused onto the target. The ejected material that arises from a target after laser irradiation is called an “ablation plume” (i.e., a plasmalike substance containing free electrons and ions, neutral particles, molecular fragments, and chemical reaction products). The physical process of laser ablation is extremely complicated, and there are several key parameters involved, such as beam energy density, the laser pulse duration, and the laser wavelength. One of the most commonly used PVD techniques for industrial applications is sputtering. The sputtering process was discovered by W.R. Grove in 1852 while studying the discharge in a tube containing gas, but its use in industry and research began in recent decades. In this process, the surface of the sputtering target is bombarded with gaseous ions under high voltage acceleration. Atoms or entire molecules of the target material are ejected and can reach the substrate. There is no melting of the material: the ejection of the particles from the source material (target) is a result of the momentum transferred from the incoming particle. The conventional setup for sputtering is a vacuum chamber where the working gas is introduced, a high negative voltage is applied to the target, e and the positively charged ionized atoms are accelerated toward it.21 23 To sputter conductive materials, direct current sputtering is used, whereas for nonconductive materials, radio frequency must be applied during the sputtering process to prevent the target from charging up due to the bombardment from positively charged ions. Another configuration is magnetron sputtering, where a magnetic field is added beneath the target to deflect and confine electrons, allowing for lower working pressures. In sputtering, the phase transition is obtained mechanically rather than chemically or thermally, so virtually any material can be deposited. The energy of the ejected molecules is higher with respect to TE and EBE, thus improving the crystallinity and adhesion of the thin film. The second group of deposition techniques is CVD: a chemical reaction transforms the molecules in the gas phase, known as the “precursors,” into solid films or powders on the substrate. There are several configurations such as , low pressure chemical vapor deposition (LPCVD), , atmospheric pressure chemical vapor deposition, One- and two-dimensional metal oxide nanostructures for chemical sensing 165

, plasma enhanced chemical vapor deposition, , photochemical vapor deposition, , laser chemical vapor deposition, and , metal organic chemical vapor deposition. In CVD, the reactant gases are diluted in carrier gases and introduced into the reaction chamber at room temperature, while the deposition surface is heated. The energy necessary to start the desired chemical reaction can be supplied as thermal energy with resistive, radiant or inductive heating, or as photon energy or glow discharge plasma. Depending on the working conditions, the reactant may experience homogeneous chemical reactions in the vapor phase before hitting the surface. Otherwise, when the reactant gases approach the surface, they slow down and heterogeneous reactions occur on the surface, forming the deposited material. Gaseous reaction by-products are then transported by the carrier gas out of the reaction chamber. Whichever heating method is employed, CVD has to provide a volatile precursor containing the elements that compose the deposited film, transport the precursor toward the substrate surface, enhance or reduce reactions in the gas phase, and provide the surface reaction needed to form the film. The setup consists of a reaction chamber; gas/vapor delivery lines; the energy source; vacuum systems (LPCVD); an exhaust system; and gas e flow, pressure and temperature monitoring systems.24 26 Hazardous vapors are also frequently used and may be produced by chemical reactions. Thus, safety equipment may be necessary. The advantages of CVD films are good adhesion, good step coverage, and high versatility of materials, but a drawback is the formation of hazardous and corrosive by-products. Beyond PVD and CVD, there are techniques from the liquid/solution phase: such as solegel, spray coating, spin coating, electrochemical deposition, and liquid phase epitaxy. The solegel process is the most widely used method for the deposition of metal oxide for gas sensors. The solegel process generally involves the transition from a liquid sol into a solid gel phase. Inorganic metal salts or metal organic compounds, such as metal alkoxides, may be used as precursors. The solegel deposition process usually has four steps: , colloidal particles are dispersed in a liquid (sol), , deposition of sol solution on the substrate by spraying, dipping, or spinning, , polymerization of the particles in the sol by stabilizing component’s removal (gel), and 166 E. Comini and D. Zappa

, heat treatment to pyrolyze the remaining organic or inorganic components forming the final film. The advantages of this process are the production of high-purity metal oxides, a highly controllable composition, the low temperature deposition, e and a simple and economic experimental setup.27 29 However, disadvantages such as weak adhesion and low wear-resistance limit its full industrial exploitation. 5.2.2 One-dimensional nanostructures A nomenclature for one-dimensional materials has not yet been established. Different and creative names have been presented in the literaturedsuch as nanotubules, -whiskers, -fibers, -fibrils, -cables, -castles, etc.daccording to the morphology on the nanostructures. Terms such as “nanowires” or “nanorods” are probably the most common in the literature for structures with two dimensions not exceeding a few hundreds of nanometers. In the past 2 decades, the number of synthesis techniques for one-dimensional nanostructures has grown exponentially. These techniques can be divided into top-down and bottom-up approaches. The first involve whittling down the size of materials from the bulk size to nanometer scale via standard microfabrication technologiesdas for example lithography, exfoliation, and lift-off processesdand allows the preparation of e well-organized nanowires.30 33 However, the crystalline quality of fabricated nanomaterials is not excellent, and the manufacturing cost for large-scale production is usually very high. The second approach, on the contrary, consists of self-assembly of atomic or molecular building blocks, by using synthesis techniques like vapor phase transport, solution-based techniques, or template growth.34 The advantages are the fine control in shape, morphology, structure, high purity, and crystallinity, together with the low cost of the experimental equipment. The main drawback is the challenging integration of the nanostructures on planar substrates, needed for the exploitation of their useful properties. Several one-dimensional oxide nanostructures with different properties and morphologies have been fabricated using bottom-up synthetic routes. Most of these structures could not have been prepared easily and economically using top-down technologies. These bottom-up techniques can be generally classified in three different types: (i) vapor phase growth methods, (ii) liquid phase growth methods, and (iii) template-assisted methods. A few morphologies of these new nanostructures with potential as chemical sensing devices are summarized schematically in Fig. 5.1. One- and two-dimensional metal oxide nanostructures for chemical sensing 167

Figure 5.1 Schematic representation of some morphologies of one-dimensional nanostructures. From left to right: nanowire, longitudinal heterojunction, core-shell heterojunction, nanotube, nanoﬁber, nanorod, and hierarchical heterostructure.

5.2.2.1 Vapor phase growth methods Among bottom-up techniques, vapor phase deposition is probably the most widely used, thanks to its simplicity and versatility, and is mainly based on a controlled condensation of a vaporized metal oxide material. To obtain one-dimensional structures, there has to be a preferential growth direction, i.e., a faster growth rate in a particular direction. Even though the exact mechanism responsible for one-dimensional growth in the vapor phase is still not clearly understood, vapor phase methods have been explored and are extensively used by many research groups to synthesize one-dimensional materials. The main advantage is its simplicity in terms of the procedure and the experimental setup used. In general, the vapor phase is obtained by evaporation of metal oxide powder (PVD), chemical reduction, or other precursor-based reactions (CVD). TE, laser ablation, or evaporation by ion, electron, and molecular beams could be used to evaporate the materials or the precursors. The vapors are then transported and condensed onto the substrate’s surface held at lower temperatures. By controlling the supersaturation of the vapor, one-dimensional materials can be easily obtained. When the growth of the nanowire crystal directly originates from the condensation from the vapor phase without the use of a catalyzer, the term ‘vaporesolid growth’ is typically used. Defect-free nanowires can be produced using this technique; however, there is still no consensus on the growth mechanism. If the growth originates from the condensation onto catalyst particles, which are liquid at such high temperatures, the growth is typically deﬁned as a “vapor liquid solid” (VLS) process. The mechanism for VLS was proposed by Wagner in 1964. Under deposition conditions, the catalyzer has to form a liquid solution with the desired material. It should also have a low 168 E. Comini and D. Zappa vapor pressure and be chemically inert. In the process, the vapor diffuses into the liquid catalyzer and, as the concentration becomes too high, the growth species precipitate to form the nanowire. The liquid phase is a preferential condensation site, and this causes a higher growth rate of the VLS with respect to the VS. Furthermore, by controlling the dimension and dispersion of the catalyzer, control can be achieved over the diameter of the nanowire. Among all vapor phase methods, the VLS process is the most successfully used and cited for generating nanowires of different oxides such as 35 36 37 38 39 40 ZnO, SnO2, In2O3, NiO, TiO2, and many more, with single- crystalline structures and in considerable amounts.

5.2.2.2 Liquid phase growth methods Wet chemistry is another widely diffused approach for fabricating one-dimensional metal oxide nanostructures. There are many experimental techniques for the preparation of nanowires from the liquid phase. A considerable research effort has been expended in developing template-free methods for the deposition of one-dimensional nanostructures in a liquid environment; the most important procedures are hydrothermal e e e methods,41 48 electrospinning,49 61 sonochemical,62 68 electrochemical anodization, and electrodeposition and surfactant-assisted synthesis.69 The hydrothermal process has been a well-known procedure for material synthesis since the 1970s. It begins with an aqueous mixture of soluble metal salt (metal and/or metaleorganic) precursors, then the solution is placed in an autoclave at a high temperature (between 100 and 300C) and under e relatively high pressure (>1 atm) conditions. ZnO nanorods,69 75 76,77 78,79 80,81 82,83 84 CuO, ceria, titania, MnO2, and Co3O4, have been prepared by using wet chemical hydrothermal approaches. Electrospinning exploits an electrical charge to force the formation of mats of fine fibers.50,85 A solid fiber is produced as the electrified jet is continuously stretched due to the electrostatic repulsions between the surface charges and the evaporation of solvent. A number of oxides have been fabricated as fibrous structures: Al2O3, CuO, NiO, TiO2, SiO2, 86e97 V2O5, ZnO, Co3O4,Nb2O5, MoO3, and MgTiO3. However, the one-dimensional nanostructures produced by electrospinning are, in general, polycrystalline. The electrochemical method is a relatively simple and effective way to prepare one-dimensional semiconductor nanostructures by anodic oxidation (anodization) or electrodeposition. In case of anodization, the metal substrate is immersed in an electrolyte solution. The substrate is One- and two-dimensional metal oxide nanostructures for chemical sensing 169 then oxidized by a controlled electric field to form porous or tubular oxide structures. The electrochemical anodization method is probably best known for the preparation of metal oxide nanotubes, but can be also used to prepare e vertically aligned nanowires.98 101

5.2.2.3 Template-assisted methods There are several references reporting on template-assisted approaches for nanofabrication such as Hulteen and Martin.102 They are regarded as one of the pioneer groups for functional nanowire array fabrication. With the use of a periodic structured template, one-dimensional nanostructures can be prepared, thanks to the conﬁnement effect of the porous template. The templates can be prepared easily with anodization. Control of the aspect ratio and the area density of one-dimensional nanostructures can be achieved by changing the diameter and length of the template and by e changing the anodization voltage.103 105 The nanostructures can be deposited into the nanopores by electrodeposition or solegel deposition methods. The advantages of being low cost and repeatable, together with their potential compatibility with silicon technologies, make these nanostructure synthesis procedures interesting. Despite its simplicity, template-based growth is characterized by the production of polycrystalline nanowires, which can limit their potential for both fundamental studies and applications.

5.3 Conductometric sensor The semiconducting properties of metal oxides are due to deviation from stoichiometry. In most oxides, such as tin oxide, oxygen vacancies are responsible for the n-type behavior.106,107 The normal working condition for a chemical sensor in the presence of air is at relatively high temperatures (500e800K). At these temperatures, the metal oxide conduction is electronic and there are ionized oxygen vacancies. Oxygen in such conditions is chemisorbed on the metal oxide surface, capturing charge carriers from the conduction band and producing a space charge area near the surface. Chemical sensing is achieved in most cases by oxidation reactions between chemical species and chemisorbed oxygen, causing a decrease in the surface barrier, leading to a change in conductance. Other chemical species, such as nitrogen oxide or water vapor, may chemisorb directly on the metal oxide surfaces by trapping or releasing electrons. 170 E. Comini and D. Zappa

5.3.1 Device integration Device integration is very easy and well-established for thin films, which may be easily patterned or deposited on the final transducers. In the case of one- dimensional nanostructures, instead, some open issues still remain.11,108 One-dimensional nanostructures should be grown directly on the transducers, but, depending on the deposition conditions, this may not always be possible due to high temperatures, pressures, or the aggressive ambient required for their preparation. In these cases, they have to be transferred afterward. The easiest way to transfer is by drop coating,109,110 but other techniques such as dielectrophoresis,111,112 or roll transfer,113 may be used, which are more compatible with industrial-scale manufacturing. Single-nanowire devices are still not exploited for mass production, due to the very precise integration process required. For such devices, nanoma- nipulation114 of the single metal oxide nanowire can be used. The problem that remains in all cases is the low mechanical and electrical stability of the contact achieved between the metal oxide and the metallic electrodes or the substrate. To obtain stable devices, which can work for very long time, there must be a good and reliable electrical contact, with the lowest contact resistance possible. This is because the metal semiconductor junction forming at the interface between the metal oxide and the metal may play a role in chemical sensing. This is even more important for single-nanowire devices, because the junction is in series with the nanowire resistance; for multiple-nanowire devices, instead, it is connected to a large number of resistances and thus less prominent. New lithographic techniques have been proposed for the integration of e the vapor phase growth process with device fabrication.115 119 Concerning chemical sensing, a high temperature lift-off procedure for the integration of a nanowire network on sensing transducers was developed by using silicon oxide as a sacrificial layer.120 This allows a clean patterning and assures the presence of uniform surfaces for the deposition of contacts. For single-nanowire devices, highly expensive techniques (such as a focused ion beam, or a series of nanolithographic tools) could be used, e ranging from proton and electron beam nanolithography,121 123 in which patterned substrates are obtained under the application of a charged particle beam, to nanoimprint lithography.124,125

5.4 Transduction principles and related novel devices When a sensing material is exposed to a specific atmosphere, it may interact with it in many different ways, which result in a change of some One- and two-dimensional metal oxide nanostructures for chemical sensing 171 of its physical properties, i.e., electrical, optical, magnetic, and even structural. Among these, electrical properties are for sure the most common and the easiest to be detected. The interaction of the sensing material with surrounding atmosphere can be transduced as a change of resistance, impedance, or work function. The easiest measurable parameter is the sensor resistance in DC conditions. It may be measured by a voltamperometric technique at constant bias but, in commercial chemical sensors, the sensing film is usually inserted inside a voltage divider. A typical kinetic response of conductance as a function of an introduction of a concentration step is shown in Fig. 5.2. After the reducing species is introduced at time t1, the sensor conductance Gi increases to Gf, in the time needed to reach the new thermodynamic equilibrium of the surface reactions. If the metal oxide is not stable, or if there is an irreversible chemisorption, a steady state may not be reached. Response time is the time necessary for the electrical conductance to reach a threshold value (usually 90%) of the difference between Gf and Gi. Recovery time is the time necessary for the conductance to recover to a level expressed as a percentage fraction (usually 90%) of GfeGi. Concern- ing the response of chemical sensors, the linearity hypothesis is not verified, and the response when working with gas mixtures cannot be deduced by the superimposition principle, with a simple sum of the individual response.

Conductance (S) Concentration (ppm)

Gf Conductance (S) Concentration (ppm)

t1 Time (S) t2 Figure 5.2 Conductance variation of the sensor produced by the introduction of a step concentration of a reducing gas. 172 E. Comini and D. Zappa

The sensor response toward a reducing species and an n-type metal oxide may be deﬁned as the relative change in conductance:

Gf =Gi For an oxidizing species and an n-type metal oxide, there is an increase in the resistance and the sensor response may be deﬁned as the relative change of resistance:

Rf =Ri The calibration curve can be obtained after measuring the response at different concentrations in the same operational conditions. The calibration curve is generally reported in a bilogarithmic scale because the relation between concentration and conductance follows a power law (Fig. 5.3). Impedance is another possible transduced signal, and it can be measured by a spectroscopic analyzer or by LCR (L ¼ inductance, C ¼ capacitance, R ¼ resistance) bridges. It may be useful to identify the different contributions to the sensor response (grain boundaries, bulk and contact) but, due to higher costs, there are no commercial devices based on this transduction. Most of the sensing performances reported in the literature are based on measurements of individual devices in artificial environments that do not reproduce field conditions. In some studies, the carrier gas is nitrogen instead of synthetic air, and no humidity or interfering gases are introduced. That is why it is very difficult, and sometimes impossible, to make a fair comparison Response

Concentration (ppm) Figure 5.3 Calibration curve of the response of a chemical sensor toward a chemical species. One- and two-dimensional metal oxide nanostructures for chemical sensing 173 of all the results reported in literature, or to speculate on sensing performances in a real environment. Few comparative studies between nanowire and polycrystalline chemical e sensors have been reported.126 129 Sysoev reported that even if the nanoparticles had a higher response to 2-propanol vapors at first, after some days of operation the response of the nanoparticles decreases to the stable response of nanowires.128 This was ascribed to the irreversible sintering process in the nanoparticles that occurs due to high temperature operation. Kumar compared different morphologies of ZnO nanostructures, and he highlighted that one-dimensional ZnO nanomaterials provide a prospective base due to their crystallinity for their applications as durable conductometric gas sensors compared with nanoparticles and thin films.129 The research on one-dimensional nanostructures is not as advanced as that on two-dimensional nanostructures, due to the difficulties in fabricating the device. Nevertheless, to exploit the unique possibilities of these structures, the focus has to be on peculiar properties that can lead to essential advances in functional devices. For example, the self-heating property can be used for the development of fully autonomous chemical sensors.130,131 Self-heating of a single nanowire is due to the dissipated power (Joule effect) induced by the bias current applied in conductometric measurements. Nanowires, with their small mass, can be heated up to several hundreds of degrees with a few tens of microwatts. Moreover, the thermal response time of these devices is extremely fast (in milliseconds range). This makes it possible to even observe the kinetics of the interactions between the gas molecules and the metal oxide. By combining low power electronics with continuous or pulsed self-heating of nanowires, it will be possible to reduce power consumption to the microwatt range, or even lower.130,131 Another most interesting approach proposed to improve chemical interactions and reduce the operating temperature is optical excitation. High temperatures limit the application of chemical sensors to nonexplosive and inflammable environments: the use of standard metal oxideebased devices at 200C or more is not recommended in presence of free hydrogen, for example. As metal oxide semiconductors absorb photons with an energy above their bandgap, free carriers are produced in the space charge area. The excess electrons are swept away from the surface, while excess holes are swept toward it due to the electrical field in the space charge area, with a decrease in the surface band bending. Several years ago, the effect of photoactivation on the sensing performances was demonstrated for 174 E. Comini and D. Zappa

e two-dimensional nanostructure metal oxide chemical sensors.132 135 The first report on the possibility of using optical excitation on one- dimensional nanostructure sensing devices was published by Law et al.136 After several years, the response of optically excited single-nanowire devices was shown to be comparable with devices that were thermally activated, in the optimal experimental conditions.137,138 Many metal oxide materials used for gas sensing applications have a wide bandgap (3.6e3.9 eV for SnO2), therefore is necessary to use UV light to excite these sensing materials. Thanks to the advances in LED fabrication, now UV LEDs (325 nm for example) are quite cheap and could be easily integrated into sensing systems.139 Conductometric devices are by far the most explored ones, but other transducing mechanisms have been investigated also. Field effect transistors (FETs) incorporating metal oxide nanowires have been fabricated, combining the advantages of conductometric devices with the possibility to further tune the sensing properties by channel modulation. These NW-FET devices have been largely used as biosensors, thanks to the possibility to functionalize the surface with specific receptors.140,141 A novel electrical transduction mechanism was recently exploited, firstly on 2D thin films142,143 and then on quasieone-dimensional metal oxide nanowires.144 Instead of measuring the relative change in the conductance of the material, surface ionizationebased devices measure the ionic current between the surface of the metal oxide material and a counter electrode, in the presence of ionized gas molecules. It was demonstrated that these devices might easily discriminate, for example, amines and hydrocarbons with amine functional groups, which enable sensors for illicit drug monitoring to be made. Optical properties are also influenced by the interaction of metal oxide surface with the surrounding atmosphere. For example, a reversible modification of static photoluminescence efficiency of ZnO nanowires was observed on exposure to low concentrations of nitrogen dioxide.145 A similar behavior was detected on TiO2/SnO2 nanoparticles on ammonia exposure.146

5.5 Conclusion and future trends Significant efforts have been made to develop and test new metal oxides, especially in the form of nanowires, nanoparticles, and nanotubes. However, their application as chemical sensors still faces problems such as One- and two-dimensional metal oxide nanostructures for chemical sensing 175 device stability over time, selectivity and long-term drift due to stoichiometry changes, and coalescence of crystallites, especially for nanoparticles. The notion of preparing multipurpose devices has been replaced by the development of sensors tailored for specific and focused applications. The improvement of computer systems and of imaging and spectroscopic techniques will provide powerful tools for the better understanding of chemical sensing mechanisms and help to optimize sensor design. One-dimensional nanostructures have a greater surface-to-volume ratio, better stoichiometry, and a higher degree of crystallinity compared with two-dimensional nanostructures. They also have reduced instability associated with grain coalescence. These factors make one-dimensional metal oxides very promising for the better understanding and the development of a new generation of chemical sensors. References 1. Brattain WH, Bardeen J. Surface properties of germanium. Bell Syst Tech J 1953;32: 1e41. https://doi.org/10.1002/j.1538-7305.1953.tb01420.x. 2. Yuliarto B, Gumilar G, Septiani NLW. SnO2 nanostructure as pollutant gas sensors: synthesis, sensing performances, and mechanism. Adv Mater Sci Eng 2015. https:// doi.org/10.1155/2015/694823. Article number 694823. 3. Galstyan V, Comini E, Faglia G, Sberveglieri G. TiO2 nanotubes: recent advances in synthesis and gas sensing properties. Sensors 2013;13(11):14813e38. https://doi.org/ 10.3390/s131114813. 4. Ilin A, Martyshov M, Forsh E, Forsh P, Rumyantseva M, Abakumov A, Gaskov A, Kashkarov P. UV effect on NO2 sensing properties of nanocrystalline In2O3. Sensor Actuator B Chem 2016;231:491e6. https://doi.org/10.1016/j.snb.2016.03.051. 5. Wang C, Sun R, Li X, Sun Y, Sun P, Liu F, Lu G. Hierarchical flower-like WO3 nanostructures and their gas sensing properties. Sensor Actuator B Chem 2014;204: 224e30. https://doi.org/10.1016/j.snb.2014.07.083. 6. Zhu L, Zeng W. Room-temperature gas sensing of ZnO-based gas sensor: a review. Sens Actuat A Phy 2017;267:242e61. https://doi.org/10.1016/j.sna.2017.10.021. 7. Tao Y, Gao Q, Di J, Wu X. Gas sensors based on a-Fe2O3 nanorods, nanotubes and nanocubes. J Nanosci Nanotechnol 2013;13(8):5654e60. https://doi.org/10.1166/ jnn.2013.7559. 8. Kim H-J, Lee J-H. Highly sensitive and selective gas sensors using p-type oxide semiconductors: overview. Sensor Actuator B Chem 2014;192:607e27. https://doi.org/ 10.1016/j.snb.2013.11.005. 9. Tonezzer M, Dang LTT, Tran HQ, Iannotta S. Multiselective visual gas sensor using nickel oxide nanowires as chemiresistor. Sensor Actuator B Chem 2018;255:2785e93. https://doi.org/10.1016/j.snb.2017.09.094. 10. Barsan N, Simion C, Heine T, Pokhrel S, Weimar U. Modeling of sensing and transduction for p-type semiconducting metal oxide based gas sensors. J Electroceram 2010; 25(1):11e9. https://doi.org/10.1007/s10832-009-9583-x. 11. Nikoobakht B, Wang X, Herzing A, Shi J. Scalable synthesis and device integration of self-registered one-dimensional zinc oxide nanostructures and related materials. Chem Soc Rev 2013;42(1):342e65. https://doi.org/10.1039/c2cs35164a. 176 E. Comini and D. Zappa

12. Yamazoe N. New approaches for improving semiconductor gas sensors. Sensor Actuator B 1991;5:7. https://doi.org/10.1016/0925-4005(91)80213-4. 13. Kanitkar P, Kaur M, Sen S, Joshi A, Kumar V, Gupta SK, Yakhmi JV. Growth and gas-sensing studies of metal oxide semiconductor nanostructures. Int J Nanotechnol 2010;7(9e12):883e906. https://doi.org/10.1504/IJNT.2010.034696. 14. Yaacob MH, Yu J, Latham K, Kalantar-Zadeh K, Wlodarski W. Optical hydrogen sensing properties of nanostructured Pd/MoO(3) films. Sens Lett 2011;9(1):16e20. https://doi.org/10.1166/sl.2011.1410. 15. Chen SE, Lu HH, Brahma S, Huang JL. Effects of annealing on thermochromic properties of W-doped vanadium dioxide thin films deposited by electron beam evaporation. Thin Solid Films 2017;644:52e6. https://doi.org/10.1016/ j.tsf.2017.05.052. 16. Wongchoosuk C, Wisitsoraat A, Tuantranont A, Kerdcharoen T. Portable electronic nose based on carbon nanotube-SnO(2) gas sensors and its application for detection of methanol contamination in whiskeys. Sensor Actuator B Chem 2010;147(2):392e9. https://doi.org/10.1016/j.snb.2010.03.072. 17. Lu C, Chen Z. High-temperature resistive hydrogen sensor based on thin nanoporous rutile TiO(2) film on anodic aluminum oxide. Sensor Actuator B Chem 2009;140(1): 109e15. https://doi.org/10.1016/j.snb.2009.04.004. 18. Durrani SMA, AI-Kuhaili MF. Effect of biasing voltages and electrode metals and materials on the sensitivity of electron beam evaporated HfO2 thin film CO sensor. Mater Chem Phys 2008;109(1):56e60. https://doi.org/10.1016/j.matchemphys. 2007.10.034. 19. Trucchi DM, Zanza A, Bellucci A, Marotta V, Orlando S. Photoconductive and photovoltaic evaluation of In(2)O(3)-SnO(2) multilayered thin-films deposited on silicon by reactive pulsed laser ablation. Thin Solid Films 2010;518(16):4738e42. https:// doi.org/10.1016/j.tsf.2009.12.072. 20. Belysheva TV, Gerasimov GN, Gromov VF, Trakhtenberg LI. The sensor properties of Fe2O3 center dot In2O3 films: the detection of low ozone concentrations in air. Russ J Phys Chem 2008;82(10):1721e5. https://doi.org/10.1134/S0036024408100142. 21. Mitu B, Marotta V, Orlando S. Multilayered metal oxide thin film gas sensors obtained by conventional and RF plasma-assisted laser ablation. Appl Surf Sci 2006;252(13): 4637e41. https://doi.org/10.1016/j.apsusc.2005.07.102. 22. Al-Hardan N, Abdullah MJ, Aziz AA. Impedance spectroscopy of undoped and Cr- doped ZnO gas sensors under different oxygen concentrations. Appl Surf Sci 2011; 257(21):8993e7. https://doi.org/10.1016/j.apsusc.2011.05.078. 23. Shaalan NM, Yamazaki T, Kikuta T. Effect of micro-electrode geometry on NO(2) gas-sensing characteristics of one-dimensional tin dioxide nanostructure microsensors. Sensor Actuat B-Chem 2011;156(2):784e90. https://doi.org/10.1016/j.snb.2011. 02.039. 24. Batista C, Teixeira V, Ribeiro RM. Pulsed DC reactive magnetron sputtering of vanadium dioxide thermochromic thin films. Mater Technol 2011;26(1):35e9. https:// doi.org/10.1179/175355511X12941605982307. 25. Haireche S, Boumeddiene A, Guittoum A, El Hdiy A, Boufelfel A. Structural, morphological and electronic study of CVD SnO2:Sb films. Mater Chem Phys 2013; 139(2e3):871e6. https://doi.org/10.1016/j.matchemphys.2013.02.046. 26. Beardslee JA, Mebust AK, Chaimowitz AS, Davis-VanAtta CR, Leonard H, Moersch TL, Afridi MY, Taylor CJ. Using precursor chemistry to template vanadium oxide for chemical sensing. Chem Vap Depos 2010;16:206e10. https://doi.org/ 10.1002/cvde.201004286. 27. Bekermann D, Rogalla D, Becker HW, Winter M, Fischer RA, Devi A. Volatile, monomeric, and fluorine-free precursors for the metal organic chemical vapor deposition of zinc oxide. Eur J Inorg Chem 2010;9:1366e72. https://doi.org/10.1002/ ejic.200901037. One- and two-dimensional metal oxide nanostructures for chemical sensing 177

28. Neri G. Non-conventional sol-gel routes to nanosized metal oxides for gas sensing: from materials to applications. Sci Adv Mater 2010;2(1):3e15. https://doi.org/ 10.1166/sam.2010.1062. 29. Breedon M, Spizzirri P, Taylor M, du Plessis J, McCulloch D, Zhu JM, Yu LS, Hu Z, Rix C, Wlodarski W, Kalantar-zadeh K. Synthesis of nanostructured tungsten oxide thin ﬁlms: a simple, controllable, inexpensive, aqueous sol-gel method. Cryst Growth Des 2010;10(1):430e9. https://doi.org/10.1021/cg9010295. 30. Sun Z, Liao T, Kou L. Strategies for designing metal oxide nanostructures. Sci China Mater 2017;60:1e24. https://doi.org/10.1007/s40843-016-5117-0. 31. Ramadan S, Kwa K, King P, O’Neill A. Reliable fabrication of sub-10 nm silicon nanowires by optical lithography. Nanotechnol 2016;27(42):425302. https://doi.org/ 10.1088/0957-4484/27/42/425302. 32. Marrian CRK, Tennant DM. Nanofabrication. J Vac Sci Technol A 2003;21:S207e15. https://doi.org/10.1116/1.1600446. 33. Candeloro P, Comini E, Baratto C, Faglia G, Sberveglieri G, Kumar R, Carpentiero A, Di Fabrizio E. SnO2 lithographic processing for nanopatterned gas sensors. J Vac Sci Technol B 2005;23:2784e8. https://doi.org/10.1116/1.2110371. 34. Zhou Y, Li J-T, Sun S-G. Synthesis-cum-assembly toward hierarchical nanoarchitectures. Coord Chem Rev 2017;352:291e305. https://doi.org/10.1016/ j.ccr.2017.09.018. 35. Serrano A, Arana A, Galdamez A, Dutt A, Monroy BM, Guell€ F, Santana G. Effect of the seed layer on the growth and orientation of the ZnO nanowires: consequence on structural and optical properties. Vacuum 2017;146:509e16. https://doi.org/10.1016/ j.vacuum.2017.03.010. 36. Terasako T, Kohno K, Yagi M. Vapor-liquid-solid growth of SnO2 nanowires utilizing alternate source supply and their photoluminescence properties. Thin Solid Films 2017;644:3e9. https://doi.org/10.1016/j.tsf.2017.05.053. 37. Tuzluca FN, Yesilbag YO, Ertugrul M. Synthesis of In2O3 nanostructures with different morphologies as potential supercapacitor electrode materials. Appl Surf Sci 2018;427:956e64. https://doi.org/10.1016/j.apsusc.2017.08.127. 38. Kaur N, Comini E, Zappa D, Poli N, Sberveglieri G. Nickel oxide nanowires: vapor liquid solid synthesis and integration into a gas sensing device. Nanotechnol 2016; 27(20):205701. https://doi.org/10.1088/0957-4484/27/20/205701. 39. Lee JC, Park KS, Kim TG, Choi HJ, Sung YM. Controlled growth of high-quality TiO2 nanowires on sapphire and silica. Nanotechnol 2006;17(17):4317e21. https:// doi.org/10.1088/0957-4484/17/17/006. 40. Klamchuen A, Suzuki M, Nagashima K, Yoshida H, Kanai M, Zhuge F, He Y, Meng G, Kai S, Takeda S, Kawai T, Yanagida T. Rational concept for designing vapor-liquid-solid growth of single crystalline metal oxide nanowires. Nano Lett 2015;15(10):6406e12. https://doi.org/10.1021/acs.nanolett.5b01604. 41. Candeloro P, Carpentiero A, Cabrini S, Di Fabrizio E, Comini E, Baratto C, Faglia G, Sberveglieri G, Gerardino A. SnO2 sub-micron wires for gas sensors. Micro Eng 2005; 78e79:178e84. https://doi.org/10.1016/j.mee.2004.12.024. 42. Liu B, Zeng HC. Hydrothermal synthesis of ZnO nanorods in the diameter regime of 50 nm. J Am Chem Soc 2003;125:4430e1. https://doi.org/10.1021/ja0299452. 43. Wang JSM, Gao L. Wet chemical synthesis of ultralong and straight single-crystalline ZnO nanowires and their excellent UV emission properties. J Mater Chem 2003;13: 2551e4. https://doi.org/10.1039/B307565F. 44. Guo M, Diao P, Cai SM. Hydrothermal growth of well-aligned ZnO nanorod arrays: dependence of morphology and alignment ordering upon preparing conditions. J Solid State Chem 2005;178:1864e73. https://doi.org/10.1016/j.jssc.2005.03.031. 178 E. Comini and D. Zappa

45. Cao MH, Wang YH, Guo CX, Qi YJ, Hu CW, Wang EB. A simple route towards CuO nanowires and nanorods. J Nanosci Nanotechnol 2004;4:824e8. https://doi.org/ 10.1166/jnn.2004.822. 46. Zhou KB, Wang X, Sun XM, Peng Q, Li YD. Enhanced catalytic activity of ceria nanorods from well-defined reactive crystal planes. J Catal 2005;229:206e12. https://doi.org/10.1016/j.jcat.2004.11.004. 47. Dharmadhikari DV, Athawale AA. Studies on structural and optical properties of rare earth copper oxides synthesized by template free hydrothermal method. Mater Sci Eng B: Solid-State Mater Adv Technol 2018;229:70e8. https://doi.org/10.1016/ j.mseb.2017.12.012. 48. Einarsrud MA, Grande T. 1D oxide nanostructures from chemical solutions. Chem Soc Rev 2014;43:2187e99. https://doi.org/10.1039/C3CS60219B. 49. Yuan ZY, Su BL. Titanium oxide nanotubes, nanofibers and nanowires. Colloids Surf A Physicochem Eng Asp 2004;241:173e83. https://doi.org/10.1016/ j.colsurfa.2004.04.030. 50. Formhals A. Process and apparatus for preparing artificial threads. 1934. US patent, 1 975 504. 51. Dai H, Gong J, Kim H, Lee D. A novel method for preparing ultra-fine alumina- borate oxide fibres via an electrospinning technique. Nanotechnol 2002;13:674e7. https://doi.org/10.1088/0957-4484/13/5/327. 52. Viswanathamurthi P, Bhattarai N, Kim HY, Lee DR, Kim SR, Morris MA. Prepara- tion and morphology of niobium oxide fibres by electrospinning. Chem Phys Lett 2003;374:79e84. https://doi.org/10.1016/S0009-2614(03)00702-4. 53. Shao C, Kim HY, Gong J, Ding B, Lee DR, Park SJ. Fiber mats of poly(vinyl alcohol)/ silica composite via electrospinning. Mater Lett 2003;57:1579e84. https://doi.org/ 10.1016/S0167-577X(02)01036-4. 54. Guan H, Shao C, Chen B, Gong J, Yang X. A novel method for making CuO super- fine fibres via an electrospinning technique. Inorg Chem Commun 2003;6:1409e11. https://doi.org/10.1016/j.inoche.2003.08.021. 55. Guan H, Shao C, Chen B, Gong J, Yang X. Preparation and characterization of NiO nanofibres via an electrospinning technique. Inorg Chem Commun 2003;6:1302e3. https://doi.org/10.1016/j.inoche.2003.08.003. 56. Yang X, Shao C, Guan H, Li X, Gong J. Preparation and characterization of ZnO nanofibers by using electrlospun PVA/zinc acetate composite fiber as precursor. Inorg Chem Commun 2004;7:176e8. https://doi.org/10.1016/j.inoche.2003.10.035. 57. Ding B, Kim H, Kim C, Khil M, Park S. Morphology and crystalline phase study of electrospun TiO2-SiO2 nanofibres. Nanotechnol 2003;14:532e7. https://doi.org/ 10.1088/0957-4484/14/5/309. 58. Viswanathamurthi P, Bhattarai N, Kim HY, Lee DR. Vanadium pentoxide nanofibers by electrospinning. Scr Mater 2003;49:577e81. https://doi.org/10.1016/S1359- 6462(03)00333-6. 59. Tavakoli Foroushani F, Tavanai H, Ranjbar M, Bahrami H. Fabrication of tungsten oxide nanofibers via electrospinning for gasochromic hydrogen detection. Sens Actuat B 2018;268:319e27. https://doi.org/10.1016/j.snb.2018.04.120. 60. Von Reitzenstein NH, Bi X, Yang Y, Hristovski K, Westerhoff P. Morphology, structure, and properties of metal oxide/polymer nanocomposite electrospun mats. J Appl Polym Sci 2016;133(33):43811. https://doi.org/10.1002/app.43811. 61. Niu C, Meng J, Wang X, Han C, Yan M, Zhao K, Xu X, Ren W, Zhao Y, Xu L, Zhang Q, Zhao D, Mai L. General synthesis of complex nanotubes by gradient electrospinning and controlled pyrolysis. Nat Commun 2015;6:7402. https://doi.org/ 10.1016/j.snb.2017.05.161. One- and two-dimensional metal oxide nanostructures for chemical sensing 179

62. Dharmaraj N, Park HC, Lee BM, Viswanathamurthi P, Kim HY, Lee DR. Prepara- tion and morphology of magnesium titanate nanoﬁbres via electrospinning. Inorg Chem Commun 2004;7. https://doi.org/10.1016/j.inoche.2003.12.033. 431e133. 63. Xu CK, Xu GD, Liu YK, Wang GH. A simple and novel route for the preparation of ZnO nanorods. Solid State Commun 2002;122:175e9. https://doi.org/10.1016/ S0038-1098(02)00114-X. 64. Xu CK, Zhao XL, Liu S, Wang GH. Large-scale synthesis of rutile SnO2 nanorods. Solid State Commun 2003;125:301e4. https://doi.org/10.1016/S0038-1098(02) 00826-8. 65. Xu CK, Xu GD, Wang GH. Preparation and characterization of NiO nanorods by thermal decomposition of NiC2O4 precursor. J Mater Sci 2003;38:779e82. https:// doi.org/10.1023/A:1021856930632. 66. Gao T, Li QH, Wang TH. Sonochemical synthesis, optical properties, and electrical properties of core/shell-type ZnO nanorod/CdS nanoparticle composites. Chem Mater 2005;17:887e92. https://doi.org/10.1021/cm0485456. 67. Miao JJ, Wang H, Li YR, Zhu JM, Zhu JJ. Ultrasonic-induced synthesis of CeO2 nanotubes. J Cryst Growth 2005;281:525e9. https://doi.org/10.1016/ j.jcrysgro.2005.04.058. 68. Kim DS, Kim JC, Kim BK, Kim DW. One-pot low-temperature sonochemical synthesis of CuO nanostructures and their electrochemical properties. Ceram Int 2016; 42(16):19454e60. https://doi.org/10.1016/j.ceramint.2016.09.044. 69. Wang Y, Liu P, Zhu K, Wang LJ. Hierarchical bilayered hybrid nanostructural arrays of NiCo2O4 micro-urchins and nanowires as a free-standing electrode with high loading for high-performance lithium-ion batteries. Nanoscale 2017;9(39): 14979e89. https://doi.org/10.1039/c7nr03979d. 70. Kumar RV, Koltypin Y, Xu XN, Yeshurun Y, Gedanken A, Felner I. Fabrication of magnetite nanorods by ultrasound irradiation. J Appl Phys 2001;89:6324e8. https:// doi.org/10.1063/1.1369408. 71. Sundararajan M, Sakthivel P, Fernandez AC. Structural, optical and electrical properties of ZnO-ZnS nanocomposites prepared by simple hydrothermal method. J Alloy Comp 2018;768:553e62. https://doi.org/10.1016/j.jallcom.2018.07.245. 72. Han SY, Akhtar MS, Jung I, Yang OB. ZnO nanoﬂakes nanomaterials via hydrothermal process for dye sensitized solar cells. Mater Lett 2018;230:92e5. https://doi.org/ 10.1016/j.matlet.2018.07.083. 73. Choi SC, Sohn SH. Controllable hydrothermal synthesis of bundled ZnO nanowires using cerium acetate hydrate precursors. Phys E Low-dimens Syst Nanostruct 2018;104: 98e100. https://doi.org/10.1016/j.physe.2018.07.014. 74. Chen X, Shen Y, Zhang W, Zhang J, Wei D, Lu R, Zhu L, Li H, Shen Y. In-situ growth of ZnO nanowire arrays on the sensing electrode via a facile hydrothermal route for high-performance NO2 sensor. Appl Surf Sci 2018;435:1096e104. https:// doi.org/10.1016/j.apsusc.2017.11.222. 75. Sun Y, Ndifor-Angwafor NG, Riley DJ, Ashfold MNR. Synthesis and photoluminescence of ultra-thin ZnO nanowire/nanotube arrays formed by hydrothermal growth. Chem Phys Lett 2006;431:352e7. https://doi.org/10.1016/j.cplett.2006.09.100. 76. Wang Y, Wang D, Yan B, Chen Y, Song C. Fabrication of diverse CuO nanostructures via hydrothermal method and their photocatalytic properties. J Mater Sci Mater Electron 2016;27(7):6918e24. https://doi.org/10.1007/s10854-016-4645-8. 77. Filipic G, Cvelbar U. Copper oxide nanowires: a review of growth. Nanotechnol 2012; 23(19):194001. https://doi.org/10.1088/0957-4484/23/19/194001. 78. Liao Y, He L, Zhao M, Ye D. Ultrasonic-assisted hydrothermal synthesis of ceria nanorods and their catalytic properties for toluene oxidation. J Environ Chem Eng 2017;5(5):5054e60. https://doi.org/10.1016/j.jece.2017.09.037. 180 E. Comini and D. Zappa

79. Karl Chinnu M, Vijai Anand K, Mohan Kumar R, Alagesan T, Jayavel R. Formation and characterisation of CeO2 and Gd:CeO2 nanowires/rods for fuel cell applications. J Exp Nanosci 2015;10(7):520e31. https://doi.org/10.1080/17458080.2013.845916. 80. Banerjee AN, Anitha VC, Joo SW. Improved electrochemical properties of morphology-controlled titania/titanate nanostructures prepared by in-situ hydrothermal surface modification of self-source Ti substrate for high-performance supercapacitors. Sci Rep 2017;7(1):13227. https://doi.org/10.1038/s41598-017-11346-2. 81. Tang Y, Ren H, Huang J. Synthesis of porous TiO2 nanowires and their photocatalytic properties. Front Optoelectron 2017;10(4):395e401. https://doi.org/10.1007/ s12200-017-0735-3. 82. Yuan ZY, Ren TZ, Du G, Su BL. A facile preparation of single-crystalline a-Mn2O3 nanorods by ammonia-hydrothermal treatment of MnO2. Chem Phys Lett 2004;389: 83e6. https://doi.org/10.1016/j.cplett.2004.03.064. 83. Wang B, Qiu J, Feng H, Wang N, Sakai E, Komiyama T. Preparation of MnO2/carbon nanowires composites for supercapacitors. Electrochim Acta 2016;212:710e21. https://doi.org/10.1016/j.electacta.2016.07.066. 84. Rui X, Tan H, Sim D, Liu W, Xu C, Hng HH, Yazami R, Lim TM, Yan Q. Template-free synthesis of urchin-like Co3O4 hollow spheres with good lithium storage properties. J Power Sources 2013;222:97e102. https://doi.org/10.1016/ j.jpowsour.2012.08.094. 85. Ngadiman NHA, Yusof NM, Idris A, Misran E, Kurniawan D. Development of highly porous biodegradable g-Fe2O3/polyvinyl alcohol nanofiber mats using electrospinning process for biomedical application. Mater Sci Eng C 2017;70:520e34. https:// doi.org/10.1016/j.msec.2016.09.002. 86. Vidyadharan B, Aziz RA, Misnon II, Anil Kumar GM, Ismail J, Yusoff MM, Jose R. High energy and power density asymmetric supercapacitors using electrospun cobalt oxide nanowire anode. J Power Sources 2014;270:526e35. https://doi.org/10.1016/ j.jpowsour.2014.07.134. 87. Archana PS, Gupta A, Yusoff MM, Jose R. Tungsten doped titanium dioxide nanowires for high efficiency dye-sensitized solar cells. Phys Chem Chem Phys 2014;16(16): 7448e54. https://doi.org/10.1039/c4cp00034j. 88. Lai D, Chen Y, Yang D, Guo W, Yu Y. Alumina (Al2O3) nanofibers from electrospinning. Adv Mater Res 2012;476e478:379e82. https://doi.org/10.4028/ www.scientific.net/AMR.476-478.379. 89. Leindecker GC, Alves AK, Bergmann CP. Synthesis of niobium oxide fibers by electrospinning and characterization of their morphology and optical properties. Ceram Int 2014;40(PB):16195e200. https://doi.org/10.1016/j.ceramint.2014.07.054. 90. Shahhosseininia M, Bazgir S, Joupari MD. Fabrication and investigation of silica nanofibers via electrospinning. Mater Sci Eng C 2018;91:502e11. https://doi.org/10.1016/ j.msec.2018.05.068. 91. Harilal M, Krishnan SG, Pal B, Reddy MV, Ab Rahim MH, Yusoff MM, Jose R. Environment-Modulated crystallization of Cu2O and CuO nanowires by electrospinning and their charge storage properties. Langmuir 2018;34(5):1873e82. https:// doi.org/10.1021/acs.langmuir.7b03576. 92. Hosseini SR, Ghasemi S, Kamali-Rousta M, Nabavi SR. Preparation of NiO nanofibers by electrospinning and their application for electro-catalytic oxidation of ethylene glycol. Int J Hydrog Energy 2017;42(2):906e13. https://doi.org/10.1016/ j.ijhydene.2016.09.116. 93. Sabzehmeidani MM, Karimi H, Ghaedi M. Electrospinning preparation of NiO/ZnO composite nanofibers for photodegradation of binary mixture of rhodamine B and methylene blue in aqueous solution: central composite optimization. Appl Organomet Chem 2018;32(6):e4335. https://doi.org/10.1002/aoc.4335. One- and two-dimensional metal oxide nanostructures for chemical sensing 181

94. Boyadjiev SI, Kéri O, Bardos P, Firkala T, Gaber F, Nagy ZK, Baji Z, Takacs M, Szilagyi IM. TiO2/ZnO and ZnO/TiO2 core/shell nanofibers prepared by electrospinning and atomic layer deposition for photocatalysis and gas sensing. Appl Surf Sci 2017;424:190e7. https://doi.org/10.1016/j.apsusc.2017.03.030. 95. Wang Y, Zhang J, Liu L, Zhu C, Liu X, Su Q. Visible light photocatalysis of V2O5/ TiO2 nanoheterostructures prepared via electrospinning. Mater Lett 2012;75:95e8. https://doi.org/10.1016/j.matlet.2012.01.074. 96. Dorneanu PP, Airinei A, Olaru N, Homocianu M, Nica V, Doroftei F. Preparation and characterization of NiO, ZnO and NiOeZnO composite nanofibers by electrospinning method. Mater Chem Phys 2014;148(3):1029e35. https://doi.org/10.1016/ j.matchemphys.2014.09.014. 97. Kanjwal MA, Barakat NAM, Sheikh FA, Khil MS, Kim HY. Physiochemical charac- terizations of nanobelts consisting of three mixed oxides (Co3O4, CuO, and MnO2) prepared by electrospinning technique. J Math Sci 2008;43(16):5489e94. https:// doi.org/10.1007/s10853-008-2835-3. 98. Varghese OK, Paulose M, Grimes CA. Long vertically aligned titania nanotubes on transparent conducting oxide for highly efficient solar cells. Nat Nanotechnol 2009;4: 592e7. https://doi.org/10.1038/nnano.2009.226. 99. Singh DP, Srivastava ON. Applied potential dependent growth of SnO2 nanostructures by anodic oxidation of tin. Adv Sci Lett 2012;16:255e60. https://doi.org/ 10.1166/asl.2012.2188. 100. Park J, Kim K, Choi J. Formation of ZnO nanowires during short durations of poten- tiostatic and galvanostatic anodization. Curr Appl Phys 2013;13(7):1370e5. https:// doi.org/10.1016/j.cap.2013.04.015. 101. Faid AY, Allam NK. Stable solar-driven water splitting by anodic ZnO nanotubular semiconducting photoanodes. RSC Adv 2016;6(83):80221e5. https://doi.org/ 10.1039/c6ra18747a. 102. Hulteen JC, Martin CR. A general template-based method for the preparation of nanomaterials. J Mater Chem 1997;7:1075e87. https://doi.org/10.1039/A700027H. 103. Chen YH, Shen YM, Wang SC, Huang JL. Fabrication of one-dimensional ZnO nanotube and nanowire arrays with an anodic alumina oxide template via electrochemical deposition. Thin Solid Films 2014;570(PB):303e9. https://doi.org/ 10.1016/j.tsf.2014.03.014. 104. Tian Y, Li Z, Dou S, Zhang X, Zhang J, Zhang L, Wang L, Zhao X, Li Y. Facile preparation of aligned NiO nanotube arrays for electrochromic application. Surf Coating Technol 2018;337:63e7. https://doi.org/10.1016/j.surfcoat.2017.12.054. 105. Su FY, Zhang WD. Fabrication and photoelectrochemical property of In2O3/ZnO composite nanotube arrays using ZnO nanorods as self-sacrificing templates. Mater Lett 2018;211:65e8. https://doi.org/10.1016/j.matlet.2017.09.085. 106. Epifani M, Prades JD, Comini E, Pellicer E, Avella M, Siciliano P, Faglia G, Cirera A, Scotti R, Morazzoni F, Morante JR. The role of surface oxygen vacancies in the NO2 sensing properties of SnO2 nanocrystals. J Phys Chem C 2008;112(49):19540e6. https://doi.org/10.1021/jp804916g. 107. Godinho KG, Walsh A, Watson GW. Energetic and electronic structure analysis of intrinsic defects in SnO2. J Phys Chem C 2009;113(1):439e48. https://doi.org/ 10.1021/jp807753t. 108. Joshi RK, Schneider JJ. Assembly of one dimensional inorganic nanostructures into functional 2D and 3D architectures. Synthesis, arrangement and functionality. Chem Soc Rev 2012;41(15):5285e312. https://doi.org/10.1039/c2cs35089k. 109. Yoshida T, Shinohara K, Itohara D, Fujita Y. Effects of thermal pressing on ZnO nanoparticle layers deposited by drop casting. e-J Surf Sci Nanotechnol 2016;14: 175e8. https://doi.org/10.1380/ejssnt.2016.175. 182 E. Comini and D. Zappa

110. Ko YH, Kim S, Yu JS. Drop-cast and dye-sensitized ZnO nanorod-based visible-light photodetectors. Phys Status Solidi Rapid Res Lett 2013;7(9):659e63. https://doi.org/ 10.1002/pssr.201307160. 111. Freer EM, Grachev O, Duan X, Martin S, Stumbo DP. High-yield self-limiting single-nanowire assembly with dielectrophoresis. Nat Nanotechnol 2010;5:525e30. https://doi.org/10.1038/nnano.2010.106. 112. Raychaudhuri S, Dayeh SA, Wang D, Yu ET. Precise semiconductor nanowire place- ment through dielectrophoresis. Nano Lett 2009;9(6):2260e6. https://doi.org/ 10.1021/nl900423g. 113. Chang Y-K, Hong FC-N. The fabrication of ZnO nanowire field-effect transistors by roll-transfer printing. Nanotech 2009;20:195302. https://doi.org/10.1088/0957- 4484/20/19/195302. 114. Ramírez FH, Tarancon A, Casals O, Rodríguez J, Romano-Rodríguez A, Morante JR, Barth S, Mathur S, Choi TY, Poulikakos D, Callegari V, Nellen PM. Fabrication and electrical characterization of circuits based on individual tin oxide nanowires. Nanotechnol 2006;17:5577e83. https://doi.org/10.1088/0957-4484/17/ 22/009. 115. Kim Y-K, Kim GT, Ha JS. Simple patterning via adhesion between a buffered-oxide etchant-treated PDMS stamp and a SiO2 substrate. Adv Funct Mater 2007;17:2125e32. https://doi.org/10.1002/adfm.200700217. 116. Maury P, Peter M, Mahalingam V, Reinhoudt DN, Huskens J. Patterned self- assembled monolayers on silicon oxide prepared by nanoimprint lithography and their applications in nanofabrication. Adv Funct Mater 2005;15:451e7. https://doi.org/ 10.1002/adfm.200400284. 117. Sun Y, Khang D-Y, Hua F, Hurley K, Nuzzo RG, Roger JA. Photolithographic route to the fabrication of micro/nanowires of III-V semiconductors. Adv Funct Mater 2005; 15:30e40. https://doi.org/10.1002/adfm.200400411. 118. Morag A, Jelinek R. “Bottom-up” transparent electrodes. J Colloid Interface Sci 2016; 482:267e89. https://doi.org/10.1016/j.jcis.2016.07.079. 119. Hong LY, Lin HN. NO gas sensing at room temperature using single titanium oxide nanodot sensors created by atomic force microscopy nanolithography. Beilstein J Nano- technol 2016;7(1):1044e51. https://doi.org/10.3762/bjnano.7.97. 120. Vomiero A, Ponzoni A, Comini E, Ferroni M, Faglia G, Sberveglieri G. Direct integration of metal oxide nanowires into an effective gas sensing device. Nanotechnol 2010;21:145502. https://doi.org/10.1088/0957e4484/21/14/145502. 121. Griffith S, Mondol M, Kong DS, Jacobson JM. Nanostructure fabrication by direct electron beam writing of nanoparticles. J Vac Sci Technol B 2002;20:2768e72. https://doi.org/10.1116/1.1526697. 122. Donarelli M, Ferroni M, Ponzoni A, Rigoni F, Zappa D, Baratto C, Faglia G, Comini E, Sberveglieri G. Single metal oxide nanowire devices for ammonia and other gases detection in humid atmosphere. Procedia Eng 2016;168:1052e5. https:// doi.org/10.1016/j.proeng.2016.11.338. 123. Morante JR. Chemical to electrical transduction mechanisms from single metal oxide nanowire measurements: response time constant analysis. Nanotechnol 2013;24(44): 444004. https://doi.org/10.1088/0957-4484/24/44/444004. 124. Mårtensson T, Carlberg P, Borgstrom M, Montelius L, Seifert W, Samuelson L. Nanowire arrays defined by nanoimprinting lithography. Nano Lett 2004;4: 699e702. https://doi.org/10.1021/nl035100s. 125. Ma P, Xu Z, Wang M, Lu L, Yin M, Chen X, Li D, Ren W. Fast fabrication of TiO2 hard stamps for nanoimprint lithograph. Mater Res Bull 2017;90:253e9. https:// doi.org/10.1016/j.materresbull.2017.03.010. One- and two-dimensional metal oxide nanostructures for chemical sensing 183

126. Sberveglieri G, Baratto C, Comini E, Faglia G, Ferroni M, Pardo M, Ponzoni A, Vomiero A. Semiconducting tin oxide nanowires and thin films for chemical warefare agents detection. Thin Solid Films 2009;517:6156. https://doi.org/10.1016/ j.tsf.2009.04.004. 127. Ponzoni A, Baratto C, Bianchi S, Comini E, Ferroni M, Pardo M, Vezzoli M, Vomiero A, Faglia G, Sberveglieri G. Metal oxide nanowire and thin films based gas sensors for chemical warefare simulant detection. IEEE Sens J 2008;8:735. https://doi.org/10.1109/JSEN.2008.923179. 128. Sysoev VV, Schneider T, Goschnick J, Kiselev, Habicht W, Hahn H, Strelcov E, Kolmakov A. Percolating SnO2 nanowire network as a stable gas sensor: direct comparison of long-term performance versus SnO2 nanoparticle films. Sens Actuat B 2009; 139:699. https://doi.org/10.1016/j.snb.2009.03.065. 129. Kumar R, Al-Dossary O, Kumar G, Umar A. Zinc oxide nanostructures for NO2 gasesensor applications: a review. Nano-Micro Lett 2014;7(2):1e24. https://doi.org/ 10.1007/s40820-014-0023-3. 130. Ramírez FH, Tarancon A, Casals O, Arbiol AJ, Rodríguez R, Morante JR. High response and stability in CO and humidity measures using a single SnO2 nanowire. Appl Phys Lett 2008;93:123110. https://doi.org/10.1016/j.snb.2006.09.015. 131. Prades JD, Hernandez-Ramírez F, Fischer T, Hoffmann M, Muller€ R, Lopez N, Mathur S, Morante JR. Quantitative analysis of CO-humidity gas mixtures with self-heated nanowires operated in pulsed mode. Appl Phys Lett 2010;97(24):243105. https://doi.org/10.1063/1.3515918. 132. Comini E, Cristalli A, Faglia G, Sberveglieri G. Light enhanced gas sensing properties of indium oxide and tin dioxide sensors. Sens Actuat B 2000;65:260. https://doi.org/ 10.1016/S0925-4005(99)00350-0. 133. Comini E, Faglia G, Sberveglieri G. UV light activation of tin oxide thin films for NO2 sensing at low temperatures. Sens Actuat B 2001;78:73e7. https://doi.org/ 10.1016/S0925-4005(01)00796-1. 134. Comini E, Ottini L, Faglia G, Sberveglieri G. SnO2 RGTO UV activation for CO monitoring. IEEE Sens J 2004;4(1):17e20. https://doi.org/10.1109/ JSEN.2003.822216. 135. de Lacy Costello BPJ, Ewen RJ, Ratcliffe NM, Richards M. Highly sensitive room temperature sensors based on the UV-LED activation of zinc oxide nanoparticles. Sens Actuat B 2008;134(2):945e52. https://doi.org/10.1016/j.snb.2008.06.055. 136. Law M, Kind H, Messer B, Kim F, Wang P. Photochemical sensing of NO2 with SnO2 nanoribbon nanosensors at room temperature. Angew Chem Int Ed 2002; 41(13):2405e8. https://doi.org/10.1002/1521-3773(20020703)41:13<2405::AID- ANIE2405>3.0.CO;2-3. 137. Prades JD, Jimenez-Diaz R, Hernandez-Ramirez F, Cirera A, Romano-Rodriguez A, Morante JR. Equivalence between thermal and room temperature UV light- modulated responses of gas sensors based on individual SnO2 nanowires. Sens Actuat B 2009;140(2):337e41. https://doi.org/10.1016/j.snb.2009.04.070. 138. Prades JD, Jimenez-Diaz R, Manzanares M, Hernandez-Ramirez F, Cirera A, Romano-Rodriguez A, Mathur S, Morante JR. A model for the response towards oxidizing gases of photoactivated sensors based on individual SnO2 nanowires. Phys Chem Chem Phys 2009;11(46):10881e9. https://doi.org/10.1039/b915646a. 139. Espid E, Taghipour F. UV-LED photo-activated chemical gas sensors: a review. Crit Rev Solid State Mater Sci 2017;42(5):416e32. https://doi.org/10.1080/ 10408436.2016.1226161. 140. Ahmad R, Mahmoudi T, Ahn M-S, Hahn Y-B. Recent advances in nanowires-based field-effect transistors for biological sensor applications. Biosens Bioelectron 2018;100: 312e25. https://doi.org/10.1016/j.bios.2017.09.024. 184 E. Comini and D. Zappa

141. Nehra A, Pal Singh K. Current trends in nanomaterial embedded ﬁeld effect transistor-based biosensor. Biosens Bioelectron 2015;74:731e43. https://doi.org/ 10.1016/j.bios.2015.07.030. 142. Hackner A, Habauzit A, Muller G, Comini E, Faglia G, Sberveglieri G. Surface ionization gas detection on platinum and metal oxide surfaces. IEEE Sens J 2009;9(12): 1727e33. https://doi.org/10.1109/JSEN.2009.2030705. 5290390. 143. Bouxin B, Maier K, Hackner A, Mueller G, Shao F, Prades JD, Hernandez- Ramirez F, Morante JR. On-chip fabrication of surface ionisation gas sensors. Sensor Actuator B Chem 2013;182:25e30. https://doi.org/10.1016/j.snb.2013.02.049. 144. Ponzoni A, Zappa D, Comini E, Sberveglieri V, Faglia G, Sberveglieri G. Metal oxide nanowire gas sensors: application of conductometric and surface ionization architectures. Chem Eng Trans 2012;30:31e6. https://doi.org/10.3303/ CET1230006. 145. Baratto C, Todros S, Faglia G, Comini E, Sberveglieri G, Lettieri S, Santamaria L, Maddalena P. Luminescence response of ZnO nanowires to gas adsorption. Sensor Actuator B Chem 2009;140(2):461e6. https://doi.org/10.1016/j.snb.2009.05.018. 146. Singh N, Pandey V, Singh N, Malik MM, Haque FZ. Application of TiO2/SnO2 nanoparticles in photoluminescence based fast ammonia gas sensing. J Opt (India) 2017;46(3):199e203. https://doi.org/10.1007/s12596-017-0404-3. CHAPTER SIX

Hybrid materials with carbon nanotubes for gas sensing

Thara Seesaard1, Teerakiat Kerdcharoen2, Chatchawal Wongchoosuk3 1Department of Physics, Faculty of Science and Technology, Kanchanaburi Rajabhat University, Muang District, Kanchanaburi, Thailand 2Department of Physics and NANOTEC Center of Excellence, Faculty of Science, Mahidol University, Ratchathewi, Bangkok, Thailand 3Department of Physics, Faculty of Science, Kasetsart University, Chatuchak, Bangkok, Thailand

Contents

6.1 Introduction 186 6.2 Synthesis of carbon nanotube 192 6.2.1 Arc discharge 192 6.2.2 Laser ablation 192 6.2.3 Chemical vapor deposition 193 6.3 Preparation of carbon nanotubedmetal oxide sensing ﬁlms 194 6.3.1 Spin-coating 194 6.3.2 Drop-coating 197 6.3.3 Screen-printing 197 6.3.4 Dip-coating 198 6.3.5 Electron beam (E-beam) evaporation 198 6.4 Sensor assembly 199 6.5 Characterization of carbon nanotubeemetal oxide materials 200 6.5.1 Raman spectroscopy 200 6.5.2 X-ray diffraction 201 6.5.3 Scanning electron microscope 203 6.5.4 Transmission electron microscopy 204 6.6 Sensing mechanism of carbon nanotubeemetal oxide gas sensors 205 6.7 Fabrication of electrodes and CNT/polymer nanocomposites for textile-based 206 sensors 6.7.1 Preparation of textile-based electrode 206 6.7.1.1 Crocheting technique 206 6.7.1.2 Embroidery technique 207 6.7.1.3 Screen printing technique 208 6.7.2 Preparation of CNT/polymer nanocomposite sensing materials 209 6.8 Sensor assembly for textile-based gas sensors 210 6.8.1 Immersion-coating technique 210 6.8.2 Drop-coating technique 211

6.9 Characterization of CNT/polymer nanocomposites sensing materials on textile 212 substrate 6.9.1 Scanning electron microscopy 212 6.9.1.1 Fabric-based embroidered gas sensors 212 6.9.1.2 Fabric-based screen-printed gas sensors 213 6.10 Sensing mechanism of CNT/polymer nanocomposites sensing materials on 215 fabric substrate 6.11 Conclusion 216 Acknowledgments 217 References 217

6.1 Introduction Until recently, classical methods such as human sensory evaluation,1 gas chromatography,2 and mass spectrometry3 have been the only available techniques for assessing the odors of objects, products, and the environment. Although the methods are reliable and accurate, practical utilization of these instruments is time-consuming, complicated, and costly. The advent of chemical gas sensors and the electronic nose (e-nose) in the 1990s4,5 has opened new opportunities for applications in many areas never seen before, especially for real-time, on-site, and rapid measurements. Since then, chemical gas sensors and the e-nose have been adopted as standard tools with which to complement, or even replace, traditional analytical instruments in many areas ranging from quality control of foods6,7 and beverages,8,9 environment protection10 to public safety.11 In general, chemical gas sensors can be classified into four types12 based on their transduction principles: optical, thermal, electrochemical, and gravimetric. Among these techniques, electrochemical transduction has so far dominated applications of chemical gas sensors in the measurement systems, because the interface setup is e more straightforward than other transduction methods.13 16 At present, most commercial chemical gas sensors adopt this technology, and metal oxide (MOX) semiconductors offer the most favored sensor architecture due to their low-cost, high sensitivity, and simplicity in function.17 One could easily combine several functional elements in the same device, such as the sensitive layer, signal converter, and control electronics. Despite the simple working principles of MOX gas sensors, the gas sensing mechanism at the microscopic level is very complex and is still not adequately understood.18,19 Gas sensors made of the same MOX materials can have different Hybrid materials with carbon nanotubes for gas sensing 187 properties depending on the fabrication techniques and preparation conditions. It is firmly believed that catalytic reduction/oxidation at the microscopic surface underlies the chemoresistive property of MOXs.20 These reactions are governed by the electronic structure, chemical composition, crystal structure, and relative orientation of the oxide surface to the analyte molecules, thereby allowing their gas sensing properties to be tuned by modifying such parameters. The most successful approach to optimizing the gas sensing properties of MOXs is to modify the microscopic structure by reducing grain size and modifying various crystallite parameters.21 Among MOX materials, tin oxides have been the most frequently used solid-state gas sensors. The sensitivity and selectivity of these materials can be tuned on the basis of structural engineering. Tin oxides have a rich set of structural parameters that can be modified. For example, tin oxide nanocrystals obtained from a spray pyrolysis experiment can have many crystallographic planes such as (110), (111), (200), (101), (011), (1,1,2), etc.22 Such crystallographic parameters are sensitive to the change in grain size responsible for the different gas sensing properties of films prepared using different conditions.23 MOXs can be doped by a small amount of metals, such as Sn, Pd, Cu, Nb, etc., to modify structural and electronic properties. It was found that doping tin oxide with Sn, In, and Nb leads to a decrease in the grain size down to the nanometer range.24 Kawamura et al. found that the interplay between different crystal growth directions can be controlled by the addition of impurities.25 Besides pure metals, MOXs can be doped or mixed with organic materials, leading to so-called “hybrid” chemical gas sensors. The field of hybrid chemical gas sensors is still in the infant stage.26 Combining both hard and soft materials into a single film is quite challenging due to complications both in the preparation and fabrication processes.27 In this chapter, we are specifically interested in the hybridized MOX gas sensors based on MOX and carbon nanotube (CNT) composites.28,29 For more information about the modification of the MOXs with other additives, other chapters in this book or the current reference section should be consulted.30 MOXs are a very robust technology mostly adopted commercially for semiconductor gas sensors. The stability and durability (an average life of 5 years) have been gladly welcomed by industries, including the environment, security, petrochemicals, and agriculture. However, MOXs have certain disadvantages that limit their applicability in many areas, such as in mobile devices where energy consumption is a major concern. Reducing the operating temperature (around 250e400 C for most MOXs) to room temperature has become a topic of research interest worldwide. 188 Thara Seesaard et al.

Because the sensing mechanism of MOXs is based on the surface reactivity of the materials to incoming analyte gases where electron transfer will play a major role,31 engineering the conductivity of the surface can lead to desirable sensing properties. Doping with impurities has been a successful technique for the modification of MOX surfaces. Apart from metal dopants as mentioned in the previous section, CNTs have several advantages over other composite materials.29 The CNT is very conductive and the gas sensing can be performed at room temperature. Consequently, mixing CNT with MOXs would result in increasing surface conductivity and reducing the operating temperature. The highly specific area of CNTs will also enhance the active surface of MOXs, leading to enhanced sensitivity and selectivity. MOX and CNT composite materials can be classified into two groups, depending on which material is the greater in the composition: MOX-decorated CNTs and CNT-doped MOXs. 1. MOX-decorated CNTs. In this case, the CNT is functionalized by attaching MOX nanoparticles, either by bonded or nonbonded interaction, onto the sidewall of the CNT.32,33 The most common method for achieving strong interaction of MOX nanoparticles on the CNT is to oxidize CNTs by strong acids to introduce carboxyl or hydroxyl groups on the CNT surface. Such functional groups can directly interact with the oxygen of the MOX nanoparticles via hydrogen bonding. The bonded interactions between such functional groups with metal atoms through the pair of electrons on the oxygen are also possible. Various MOX nanocrystals coated onto CNTs have been investigated for their 34 35 36 37 38 functionality, such as ZnO, TiO2, SiO2, SnO2, MnO2, and 39 Fe2O3. Such functionalization creates novel properties which extend the applicability of CNT into many new areas, such as capacitors, photocatalysts, and batteries. In gas sensing applications, MOX- decorated CNTs have shown enhanced sensitivity, improved response and recovery times, and a dramatic reduction in operating temperature. Several ambient gases have been reporteddfor example, CO, NO2, NH3, and ethanol (see reference 29 and comprehensive reference list therein). 2. CNT-doped MOXs. In this case, CNTs are embedded within the MOX matrix. This chapter will focus on this type of CNT and MOX hybrid material. The MOX/CNT thin films can be prepared using various techniques, such as spin-coating, drop-coating, dip-coating, and electron beam evaporation, details of which will be given in the following sections. Hybrid materials with carbon nanotubes for gas sensing 189

Although the MOX hybrid materials were actively developed by many researchers for the last several decades, but in fact the MOX gas sensors have certain restrictions that do not support the development in wearable sensing technologies. A desire to overcome these restrictions has given impetus to researchers to focus on searching for new gas sensing materials and emerging technologies in the field of textile-based gas sensors. CNT/polymer nanocomposite materials have received special attention from researchers worldwide, as it can operate at room temperature, low power consumption requirements, and offer several design possibilities with the e-textiles.40 In addition, CNT/polymer nanocomposite materials are also investigated as alternative gas sensing materials for several applications such as healthcare,41 e agriculture,42 industry,43 and environment.44 46 Recently, researchers have developed a wearable smart technology to change both a design and a fabrication process by using electronic circuit and digital components embedded in clothing known as e-textile or electronic textile. E-textile technology revolts fiber and textile manufacturing and has special properties that cannot be found in a traditional fabric such as in communication, perceptive functions, and conduct energy.47,48 An example of wearable e-textile innovation resulting from the combination of textile and electronic technology providing an ideal platform for wearable health tech devices is the biometric smart shirt; a body metric system that monitors cardiac, breathing patterns49 and body activity tracker of baby sleep positions with several sensors embedded in the fabric.50 There are also smart sock track cadences which can access to real-time biometric data to indicate walking abnormalities, gait analysis, and weight distribution of the body evenly around the foot while walking and running.51 Consequently, we will be able to track the progress of effectiveness of exercise. However, it might be possible to integrate electronic components into textiles or fabrics in the early stages of e-textile technology because the device connectivity component is unfashionable due to untidiness of the electrical wiring and inflexibility of the cables when woven into clothing. It is likely that hazard may occur especially while wearing an unwieldy and inconvenient garment. Then, the researchers try to figure out the appropriate method to merge electronics with clothing and jewelry. As a result, these devices will be safe to use and real-time application monitoring will be easily built. It is possible to have a more perfect form of communication without any cables due to the wireless platform for new frontier of e-textile applications such as e-textile fashion, finishing and decoration, military applications, and smart clothing.52 190 Thara Seesaard et al.

In addition, there are new developments in electrical properties of textile materials, including a conductive thread, metallic finish yarn, silver yarn, polyester powder coating/metal finishes, and synthetic fabrics consisting of fibers with high electrical conductivity.53 The innovative e-textile devices and textile materials (fabrics, yarns, and threads) have been designed with different types of fabrication techniques and processes such as luminous fabric and light-emitting fiber which were developed to decorate the building and automotive parts.54 Besides, Philips research and the Institute of Textile Institute TITV Greiz, Germany, has collaborative learning in developing e-textile technology for health and quality of life, which is known by the name of photonic textiles. Photonic textiles were created by embedding LEDs into plastic or film and then woven into the fabric to make them soft and flexible to allow unlimited shapes, which can increase the user interaction by combining the functionality of sensors and communication devices.55 Additionally, PLACE-it Project (Platform for Large Area Conformable Electonics by Integration) has developed an optoelectronic which is extremely thin, lightweight, and can be easily applied for using in health and medicine, with some applications, such as skin treatments and measuring the circulation of the blood from all parts of the body. Moreover, it can also be used in product design such as lamps, curtains, advertising, and fashion.56 E-textile can not only integrate electronics directly into the textile substrates but also make electronic components from fibers and textiles. For example, researches from NC STATE UNIVERSITY College of Textiles have demonstrated that the development of lithium-ion battery provides better performance by using MnOx/C nanofibers instead of graphite in the anode of the battery, called 18650 cells.57 An interesting smart textile project, the ProeTEX (PROtection E-TEXtiles), is the research cooperation of Euro- pean countries which gives greater importance to the development of e-textile-based MicroNano technology and wearable system for rescue workers and firemen.58,59 Recent research in electronic textile technology field has found out that the key component of this technology is the electronic circuit part on textile for controlling and processing equipment. Thus, new fabrication process and materials in the development of electronic component on textiles substrate providing a consistent quality of the e-textile work piece in every production is a core basis for creating a wearable electronic textile system. Currently, there has been increasing interest in using screen printing technology as a manufacturing process in Hybrid materials with carbon nanotubes for gas sensing 191 the development of electronic commerce research to construct a perpendicular wiring structure on textile substrate,60,61 which is possible to produce electronic components and circuits on textile that cannot be achieved by other methods such as wet chemical or photolithographic technology. Electronic circuit board in industrial electronic sector were developed by using a screen printing process for the reason that this process can produce many items in a relatively short period of time and they can be reproduced at low cost with regular quality of the work piece and precise pattern control. Although the e-textiles were actively developed by many researchers in the last decade, the innovative e-textile products were found mainly in fashionable clothing and decoration. While there are a lot of researches to support learning and creating innovation-oriented research for healthcare applications most of which are the innovations that can aid interpretation of measurement resulting in terms of physiological parameters and biokinetics such as respiration, movement, touch, brain waves, heart rate, breathing, cardiac activity, and body temperature.62 It can be seen that among a variety of smart textile innovations, there is still space available for the creation of an innovative e-textile for molecular detection using nanomaterial and nanohybrid materials. To make e-textile technology more perfect, researchers have been conducting research, which is related to the development of textile-based gas sensors used as wearable electronic noses.63 Research has been carrying on in the direction of e-textile development so as to be widely accepted. Sniffing e-textile innovation was designed to be more fashionable and comfortable for using in daily life so that the wearer would be able to work beyond the original limit. At first, the circuit board was developed from a rigid material and then it was changed to soft material such as fabric, rubber, and plastic to make it flexible. In addition, fabric substrate is also friendly to human skin and esthetically acceptable. For researchers, the biggest challenge is to overcome those restrictions. The details of research are related to the e-textile technology and clothing is the first priority that we emphasize on because it is what you can wear throughout your whole life. Moreover, clothing can not only be easily adapted to biological function that actually happens but also give comfort to the physical mobility. In this chapter, we will highlight the most important fabrication process of flexible CNTebased textile gas sensors, characteristics of CNT/polymer nanocomposites for multifunctionality sensing, and the main progress in gas sensing. 192 Thara Seesaard et al.

6.2 Synthesis of carbon nanotube CNTs are generally produced using three main techniques: arc discharge, laser ablation, and chemical vapor deposition (CVD). Each technique can be modified to suit the specific research purpose. 6.2.1 Arc discharge Arc discharge was the first technique recognized for producing multiwalled carbon nanotubes (MWCNTs)64 and single-walled carbon nanotubes (SWCNTs).65,66 The arc discharge technique generally involves the use of two high-purity graphite electrodes as anode and cathode. The electrodes are vaporized by the passage of a DC current (w100 A) through thetwohigh-puritygraphiteelectrodesseparated(w1e2mm) in 400 mbar of helium atmosphere. After arc discharging for a period of time, a carbon rod is built up at the cathode. The native method will mainly produce MWCNTs, rather than SWCNTs. However, with the addition of a metal catalyst, such as Fe, Co, Ni, Y, or Mo, on either the anode or the cathode, SWCNTs can also be produced. The quantity and quality (such as length, diameter, purity, etc.) of the nanotubes obtained depend on various parameters, such as the surface density of the metal catalysts, inert gas pressure, type of gas, plasma arc, temperature, current, and system geometry. 6.2.2 Laser ablation Smalley and colleagues produced CNTs using the laser ablation technique in 1995.67 For the laser ablation technique, a high-power laser is used to vaporize carbon from a graphite target at high temperature. Both MWCNTs and SWCNTs can be produced with this technique. To generate SWCNTs, metal particles must be added as catalysts to the graphite targets, similar to the arc discharge technique. The quantity and quality of CNTs produced depend on several factors, such as the amount and type of catalysts, laser power and wavelength, temperature, pressure, type of inert gas, and the fluid dynamics near the carbon target. The laser was focused onto a carbon target containing 1.2% of cobalt/nickel with 98.8% of graphite composite that was placed in a 1200 C quartz tube furnace under an argon atmosphere (w666.61 mbar). These conditions were achieved for production of SWCNTs in 1996 by Smalley’s group.68 In such a technique, argon gas carries the vapors from the high temperature chamber into a cooled collector positioned downstream. The nanotubes will self-assemble from Hybrid materials with carbon nanotubes for gas sensing 193 carbon vapors and condense on the walls of the flow tube. The diameter distributions of SWCNTs that result from this method vary by about 1.0e1.6 nm. CNTs produced by laser ablation were purer (up to 90% purity) than those produced by the arc discharge process and have a very narrow distribution of diameters. 6.2.3 Chemical vapor deposition The use of the CVD technique to produce MWCNTs was first reported by Endo and his research group in 1993.69 Three years later, Dai in Smalley’s group successfully adapted CO-based CVD to produce SWCNTs.70 The CVD technique can be achieved by taking a carbon source in the gas phase and using an energy source, such as plasma or a resistively heated coil, to transfer energy to a gaseous carbon molecule. The CVD process employs hydrocarbons as the carbon sources, including methane, carbon monoxide, and acetylene. The hydrocarbons flow through the quartz tube placed inside an oven at a high temperature (w720 C). At high temperature, the hydrocarbons are broken down to hydrogen and carbon radicals, producing pure carbon clusters. The carbon then diffuses to the substrate, which is heated and coated with a catalyst (usually a first-row transition metal such as Ni, Fe, or Co) where CNTs will be formed if the proper parameters are maintained. The advantages of the CVD process are low power input, lower temperature range, relatively high purity, and, most importantly, the possibility of scaling up the process. This method can produce both MWCNTs and SWCNTs depending on the temperaturedproduction of SWCNTs will occur at a higher temperature than MWCNTs. It should be noted that SWCNTs can be classified into metallic and semiconducting types depending on their diameters and chiralities. Synthesis of SWCNTs always produces a mixture of metallic and semiconducting SWCNTs which is one of the crucial problems for the development of SWCNT-based electronic applications. Nowadays, several methods, including amine extraction,71 DNA separation,72 modified free solution electrophoresis,73 density- gradient ultracentrifugation,74 polymer wrapping,75 etc., have been employed to separate semiconducting and metallic SWCNTs. The most recent progress on the structure separation of SWCNTs can be found in a literature.76 In this chapter, most of reviewed papers did not mention to the type of SWCNTs. Therefore, it can be assumed to a mix of semiconducting and metallic types and effects of the types on gas sensing properties are neglected. 194 Thara Seesaard et al.

6.3 Preparation of carbon nanotubedmetal oxide sensing films Sensing film is the heart of a gas sensor device. The key to success in developing a gas sensor device is a technique capable of preparing a sensing film that exhibits high selectivity and sensitivity to a desired target gas, long-term stability, good repeatability, rapid response, small size, and low power consumption. Until now, it has been widely known that most commercially available gas sensors are still based on pure MOX gas sensors (i.e., SnO2 and WO3). Such gas sensors have been successfully used in many applications, but they still suffer from poor selectivity and high power consumption. These disadvantages can be crucial obstacles for the development of future advanced technology, such as wearable sensing devices. Recently, doping of CNTs into MOX has attracted considerable attention because hybrid SWCNTs/SnO2 sensors exhibit high sensitivity and a good 77 recovery property in detecting NO2 at room temperature. The hybrid CNT/MOX gas sensors based on thin-film nanostructures are summarized in Table 6.1. As shown in Table 6.1, there are five methods to deposit hybrid CNTs/ MOXs onto electrodes. 6.3.1 Spin-coating Spin-coating is a method for applying liquid-based coatings onto a rotating substrate. A typical spin-coating process consists of four basic stages,89 as shown in Fig. 6.1. The coating liquid material is applied to the top of substrate in the deposition stage. The amount of applied liquid depends on the viscosity of the liquid and the size of the substrate to be coated. In the acceleration stage, liquid is spread across the wafer by centrifugal force. The spinning speed is set at a specific value depending on the desired film thickness. The coated substrate is then spun at a higher speed. The liquid flows radially outward, whereas excess liquid flows to the perimeter and leaves as droplets. In the final stage, evaporation of the solvent takes over as the primary mechanism of thinning. The thickness of the dry film (Lfilm) with an approximation of constant evaporation and no liquid remaining at the end of the process can be written as follows:90,91 " # = 0 I e 1 3 3b x x k = L ¼ 0 A A 1 x0 u 1 2 (6.1) film 2r A yrdmtraswt abnnntbsfrgssensing gas for nanotubes carbon with materials Hybrid

Table 6.1 List of hybrid CNT/MOX gas sensors. Target gas Detection range Operating temperature Sensing material Fabrication technique References NO2 25e1000 ppm 25 C SWCNTs-SnO2 Spin-coating 77 NO2 100e500 ppb 25 C MWCNTs-SnO2 Drop-coating 78 CO 10e50 ppm 150 C MWCNTs-WO3 Drop-coating 78 CH2O 0.03e10 ppm 250 C MWCNTs-SnO2 Screen-printing 79 NH3 60e800 ppm 25 C MWCNTs-SnO2 Spin-coating 80 EtOH/MeOH 100e1000 ppm 250 C MWCNTs-SnO2 E-beam evaporation 81 H2 5000e50,000 ppm 250 C MWCNTs-WO3 E-beam evaporation 82 NO2 100e1000 ppb 25 C MWCNTs-WO3 Drop-coating 83 NH3 20 ppm 25 C CNTs-ZnO Spin-coating 84 CH3COCH3 1 vol.% 25 C MWCNTs-TiO2 Screen-printing 85 NH3 1 vol.% 25 C MWCNTs-TiO2 Dip-coating 86 O2 10 ppm 350 C CNTs-TiO2 Drop-coating 87 H2 4% in air 25 C SWCNTs-Co3O4 Spin-coating 88 195 196 Thara Seesaard et al.

(a) (b)

dω ≠ dt 0

ω (c)ω (d)

Figure 6.1 The four basic stages of spin coating: (a) deposition, (b) acceleration, (c) ﬂow domination, and (d) evaporation.

cD p M ek ¼ g A A (6.2) rb1=2 RT g b 0 where 0 is kinematic viscosity; xA represents the initial concentration of I solvent in the coating liquid; xA represents the mass fraction of solvent in the coating liquid that would be in equilibrium with the mass fraction of solvent in the bulk gas; u is the spin speed; c denotes the ratio of kinematic viscosity and mass diffusivity of the ambient gas; Dg is the binary diffusivity of the solvent in the ambient gas; pA is the vapor pressure of the pure solvent at temperature (T); MA is the molecular weight of the solvent; and R, r, and bg denote the universal gas constant, liquid density, and kinematic viscosity of the ambient gas, respectively. To prepare the CNT/MOX liquids for spin-coating, the SWCNT bundles were dispersed in the organometallic solutions (Sn [OOCCH (C2H5)-C4H9]2,aq; tin (II) 2-ethylhexanoate w90% in 2-ethylhexanoic acid)77 by ultrasonic vibration. Alternatively, MWCNT bundles with SnO2 nanoparticles and cetyltrimethyl ammonium bromide can be dispersed in water.80 In the case of synthesis of CNT-ZnO,84 CNT in a mixture of ethanol and water was dropped into triethanolamine and 88 ZnCl2 solution at a specific temperature. For SWCNT/Co3O4 thin films, these can be formed by spin-coating a metalepolymer complex (Cox(C2H5N)n), as a product of the reaction of CoSO4$7H2O and polye- thylenimine in water, onto an SWCNT thin film. After deposition, the Co3O4 and SWCNT thin films were annealed at a high temperature to form Co3O4/SWCNT composite. Hybrid materials with carbon nanotubes for gas sensing 197

6.3.2 Drop-coating Drop-coating is a simple method for preparing films and employs a precision pipette to put solution onto the substrate. The film thickness is controlled by the amount and concentration of solution deposited on the substrate. This technique is closely related to spin-coating. Therefore, it can be used as an alternative when spin-coating is not possibledfor example, when the solvents are not sufficiently volatile to evaporate during a spin-coating process or not sufficiently viscous to produce a thick film.92 Preparation of some CNT/MOX sensors78,83,87 successfully uses this method for depositing films. Before a drop-coating process, hybrid CNT/MOX solution is usually prepared using an adapted solegel method for obtaining well-dispersed CNT in the MOX matrix. For example, CNT/TiO2 was obtained by adding CNTs to TiO2 prepared from titanium isopropoxide(IV) Ti [OCH(CH3)2]4 precursors in a dry nitrogen atmosphere. An adequate mixture of the two components was obtained by dissolving them in glycerol (employed as an organic vehicle) and stirring the resulted solution in an ultrasonic bath at a specific temperature.78 6.3.3 Screen-printing Screen-printing is a commonly used industrial technique for fast and inexpensive deposition of films over large areas. The principle of screen- printing93 is shown in Fig. 6.2.

Squeegee Paste Screen

Snapp-off Emulsion Substrate

Printing

Levelling

Wet film Figure 6.2 The screen printing process. 198 Thara Seesaard et al.

A pattern is photographically defined on a stainless steel screen by means of an emulsion layer. A paste of the material to be screen-printed is pressed through the screen using a foam applicator (squeegee). After leveling, the printed wet film is dried at a specific temperature. The thickness of the screen-printed film depends on the viscosity of the paste, the pressure and speed of the squeegee, the snap-off distance between the screen and the substrate, and the mesh number of the screen. In a roller squeegee system, Fox and his colleague employed a numerical model to estimate deposition thickness at different half-tone coverage (more detail can be found in Ref. 94). Preparation of a paste to fabricate an MWCNT/TiO2 gas sensor 85 was reported by Sanchez et al. The MWCNT/TiO2 composites were prepared by solegel techniques using titanium tetraisopropoxide [Ti(C3H6OH)4] as the precursor and 2-propanol as the solvent. The mixture was added in HCl and heated at a specific temperature. The solid output was mixed with few drops of Triton-X and propylene glycol to prepare a paste for the screen-printing method. 6.3.4 Dip-coating Dip-coating can be described as a process where a substrate is dipped into a solution. It is then withdrawn from the solution at a controlled speed under controlled temperature and atmospheric conditions. The coating thickness is primarily affected by the withdrawal speed, fluid viscosity, fluid density, and surface tension. If the withdrawal speed is chosen such that the sheer rates keep the system in the Newtonian regime, the coating thickness (LDip) can be calculated by the LandaueLevich equation:95,96 ðhnÞ2=3 LDip ¼ 0:94 (6.3) g1=6ðr Þ1=2 LV g where h denotes fluid viscosity, v represents the withdrawal speed, gLV is the liquidevapor surface tension, r is fluid density, and g is gravity. 86 In the case of MWCNT/TiO2 gas sensors, Sanchez and Rincon employed dip-coating based on a solegel solution. The solegel solution containing Ti-isopropoxide and acid-treated MWCNTs was either precipitated or kept as a sol by adjusting the pH and surfactant concentration. 6.3.5 Electron beam (E-beam) evaporation The electron beam (E-beam) evaporation process is a physical vapor deposition that yields a high deposition rate from 0.1 to 100 mm/min at relatively low substrate temperatures. The E-beam process offers extensive possibilities Hybrid materials with carbon nanotubes for gas sensing 199 for controlling film structure and morphology, with desired properties such as dense coating, high thermal efficiency, low contamination, high reliability, and high productivity. The deposition chamber is evacuated to a pressure of 1.33 10 5 mbar or lower. The material to be evaporated is in the form of ingots or a compressed solid. The E-beam can be generated from electron guns by thermionic emission, field electron emission, or the anodic arc method. The electron beam is accelerated to a high kinetic energy and focused toward the starting material. The kinetic energy of the electrons is converted into thermal energy that will increase the surface temperature of the materials, leading to evaporation and deposition onto the substrate. The deposition rate depends on the starting material and E-beam power. The deposited film thickness can be measured in situ by a quartz crystal monitor. The evaporation of CNTs with MOXs (i.e., SnO2 and WO3) is a relatively new concept. A plausible mechanism for CNT/ MOX coevaporation can be drawn as follows81,82: the E-beam is used to bombard the surface of the starting materials (i.e., CNT/SnO2 or CNT/ WO3). The MOXs (such as SnO2 or WO3) are evaporated at a temperature of w1500 C in a high vacuum, while CNT fragments that are small and very light are carried into the vapor by surrounding SnO2 or WO3 molecules. It should be noted that CNTs themselves are not decomposed during evaporation because this temperature is well below CNT sublimation point (>3000 C) in a high vacuum condition. When CNT molecular fragments arrive at the substrate, SnO2 or WO3 vapor is condensed and coated around them. As the substrate cools down, CNTs remain in the lattice of MOXs due to physicochemical binding between the MOXs and CNTs.

6.4 Sensor assembly A typical sensor structure is displayed in Fig. 6.3. The sensing ﬁlm is deposited on top of a substrate between the electrodes. The heater is also integrated on the reverse of the substrate. It should be noted that a heater unit may not be necessary if the sensor will be operating at room temperature.

Sensing Electrode material Electrode

Substrate

Heater

Figure 6.3 A simple sensor structure. 200 Thara Seesaard et al.

Apart from the sensing ﬁlm, the electrodes also play an important role in gas sensing response. For instance, the electrode material, gap sizes, and e electrode structure can affect the sensor response.97 99 Mishra and Agar- 97 wal reported that the sensitivity of the thick-ﬁlm SnO2 sensor for H2 and CO is much higher when silver electrodes are used instead of gold electrodes (about 65.5% and 42.6%, respectively). Tamaki et al. found that sensitivity was increased with decreasing gap size.98 The performance of the sensor was improved by using interlacing electrodes.99 Therefore, the design of a gas sensor structure is necessary for fabricating a high- performance hybrid CNT/MOX sensing device.

6.5 Characterization of carbon nanotubeemetal oxide materials To conﬁrm the structure and quality of produced CNT and MOX hybrid materials, there are four characterization techniques that are normally used. These techniques are described in the following subsections. 6.5.1 Raman spectroscopy Raman spectroscopy is a spectral measurement based on inelastic scattering of monochromatic radiation. When a molecule is irradiated with an intense monochromatic light (usually a laser source), photons excite the molecule from the ground state to a virtual energy state. The photons are reemitted when the molecule relaxes. The frequency of the reemitted photons shifts in comparison with the original monochromatic light frequency. This shift provides information about vibrational, rotational, and other low frequency transitions in molecules. Information from Raman spectroscopy is summarized in Fig. 6.4.

Characteristic Changes in Analysis raman frequency of Polarization of Width of Intensity of frequencies raman peak raman peak raman peak raman peak

Crystal Composition of Stress/strain Qulaity of Amount of Properties symmetry and material state crystal material orientation Figure 6.4 Information from Raman spectroscopy. Hybrid materials with carbon nanotubes for gas sensing 201

G-band

CO O 3 4 D-band Intensity (a.u.)

400 800 1200 1600 Raman shift (cm–1)

Figure 6.5 Raman spectra of single-walled carbon nanotube/Co3O4 film (upper line), 88 Co3O4 thin film (middle line), and the SiO2/Si substrate (lower line). Raman spectroscopy was used to confirm the existence of CNTs in an 88 MOX film. Raman spectra of an SWCNT/Co3O4 film are displayed in Fig. 6.5. The peak of crystalline Co3O4 can be clearly observed at 1 694 cm for the A1g mode, while it appears as two significant peaks for SWCNTsdnamely D-band and G-band at 1350 and 1590 cm 1, respectively. It should be noted that the intensity of the D-band (w1300e1500 cm 1) is a qualitative metric of SWCNT defects holding significant information on the crystalline quality, while the G-band (w1500e1605 cm 1) is derived from the in-plane vibration usually existing in graphite and useful for measuring SWCNT graphene sheet folding. For analysis of a CNT/MOX sensing film, the Raman shift for the MOXs (i.e., SnO2,WO3, TiO2, etc.), D-band, and G-band should be observed. 6.5.2 X-ray diffraction X-ray is a high-energy electromagnetic radiation having energies ranging from w200 eV to 1 MeV. The X-ray diffraction (XRD) is based on the elastic scattering of monochromatic X-rays. It is usually used to characterize the chemical composition and crystallographic structure of materials by plotting the angular positions and intensities of the resultant diffracted peaks of radiation satisfied with Bragg’s law conditions. The diffraction intensity can be written as follows:100 3 2 2 2 2 I0l e MðhklÞ 2 1 þ cos ð2qÞcos ð2qmÞ na Ið Þa ¼ Fð Þa hkl p 2 2 hkl 2 q q m 64 r mec Va sin cos hkl s (6.4) 202 Thara Seesaard et al.

where I(hkl)a is the intensity of the reflection of hkl in phase a, I0 is the incident beam intensity, l denotes the X-ray wavelength, r denotes the 2 2 2 distance from the specimen to the detector, (e /mec ) represents the square of the classical electron radius, M(hkl)a is the multiplicity of reflection of hkl in phase a, Va is the volume fraction of phase a, F(hkl)a is the structure factor for reflection hkl of phase a (i.e., the vector sum of scattering intensities of all atoms contributing to that reflection), 2qm represents the diffraction angle of the monochromator, va is the volume of the unit cell of phase a, and ms is the linear absorption coefficient of the specimen. 80 The XRD patterns of MWCNT/SnO2 are shown in Fig. 6.6.In general, an XRD pattern of CNT locates near the (002), (100), (110), and (112) reflections of graphite. The prominent peak (2q z 26) can be attributed to the (002) reflection of carbon. In this case, the most intense two peaks of MWCNTs correspond to (002) and (100), while only SnO2 in the crystalline phase can be indexed from the patterns for SnO2. It can be observed that the characteristic peaks of MWCNT/SnO2 composites are quite similar to the patterns of SnO2. From this observation, it may be hy- pothesized that the MWCNTs are well-embedded in the SnO2 matrix or there are no MWCNTs in the SnO2 matrix. However, almost all CNT/ MOX films from other studies have a similar pattern. Peaks of CNT are usually absent for the CNT/MOX composite films in the XRD analysis. Other techniques may need to confirm the existence of CNTs in MOX films.

Figure 6.6 X-ray diffraction patterns of (a) SnO2, (b) multiwalled carbon nanotubes 80 (MWCNTs), and (c) SnO2/MWCNTs composites. Hybrid materials with carbon nanotubes for gas sensing 203

6.5.3 Scanning electron microscope The scanning electron microscope (SEM) employs a focused beam of high- energy electrons to generate a variety of signals at the surface of sample. The types of signals produced from the interaction of the high-energy electrons with the sample include secondary electrons, back-scattered electrons, characteristic X-rays, and other photons of various energies. These signals can be used to examine many characteristics of the samples, such as surface topography and morphology and crystallographic information and composition. The basic principle of SEM is shown in Fig. 6.7. The SEM surface morphology of a CNT/WO3 film prepared by E-beam evaporation is displayed in Fig. 6.8. One can see that the sensing film prepared by this technique is highly homogeneous, with grain sizes ranging from 40 to 80 nm. It should be noted that the surface morphology of other films (including pure SnO2, pure WO3, and CNT/SnO2) prepared by E-beam evaporation81,82 is in accordance with observations on the nanocrystalline CNT/WO3 film. With SEM resolution, a CNT structure cannot be observed on the thin film surface. In cases of CNT/MOX films prepared by other methods (i.e., spin-coating),77 the morphologies of the pure MOX

Figure 6.7 Principle of scanning electron microscope. 204 Thara Seesaard et al.

Figure 6.8 Scanning electron microscopic image of multiwalled carbon nanotubese

WO3 thin film on Si substrate. and hybrid CNT/MOX are also very similar. Thus, it is quite difficult to observe the CNTs on the surface. In the previous studies, it was suggested that CNTs are mostly embedded in the MOX-based matrix. 6.5.4 Transmission electron microscopy Transmission electron microscopy (TEM) provides a much higher spatial resolution than SEM. TEM can facilitate study of the inner structure and analysis of the features on an atomic scale (in the range of a few nanometers). Although the TEM technique involves electrons to produce enlarged images similar to the SEM technique, the working principle of TEM is somewhat different from SEM. In general, TEM uses high E-beam energies in the range of 60e350 keV to pass through a thin sample to project an image onto a fluorescent screen. The sample for TEM is usually required to be sliced into an extremely thin section (<100 nm) and pretreated with heavy metals (staining) before visualization. The image resolution of TEM (d)in terms of the classic Rayleigh criterion for the visible-light microscope can be given approximately by Eq. (6.5): 0:61l d ¼ (6.5) m sin b where l denotes the wavelength of the radiation, m represents the refractive index of the viewing medium, and b is the semiangle of collection of the magnifying lens. Hybrid materials with carbon nanotubes for gas sensing 205

Figure 6.9 Typical high-resolution transmission electron microscopic image of (a) CNT/

WO3 ﬁlm and (b) nanocrystalline WO3 prepared by E-beam evaporation technique.

TEM characterization can be used to confirm CNT inclusion in MOX films. A typical high-resolution TEM (HRTEM) image of a CNT/WO3 composite prepared by E-beam evaporation is shown in Fig. 6.9. It should be noted that HRTEM uses both transmitted and scattered beams to create an interference image. HRTEM observation clearly shows that a single MWCNT fragment is, indeed, embedded into the nanocrystalline WO3 layer (see Fig. 6.9(a)). The diameter of CNTs and the crystal size of WO3 were estimated to be in the range of w20e50 nm (see Fig. 6.9(a)) and 3e10 nm (see Fig. 6.9(b)), respectively. By comparison with pure WO3, the doping of CNT does not change the phase or surface morphology of the film, but it may help form nanochannels in the MOX films, leading to the enhancement of the sensitivity and reduction of the operating temperature.

6.6 Sensing mechanism of carbon nanotubeemetal oxide gas sensors

It is widely known that many MOXs (such as WO3, SnO2, and TiO2) are n-type semiconductors, while CNT is a p-type semiconductor. CNT/ MOX gas sensors can be either p-type or n-type semiconductors, depending on the quantity of CNTs and the operating temperature. The CNT/MOX gas sensor behaves as an n-type semiconductor if the electrical conductivity of the gas sensor increases when reducing gases (i.e., H2, CO, or NH3) are absorbed by its surface. In the case of the p-type semiconductor, the electrical conductivity of the sensor increases in the presence of an oxidizing gas 206 Thara Seesaard et al.

(O3,NOX, etc.). Various oxygen species chemisorbed at the surface (such as 2 O ,O2 , and O ) are available for catalytic reactions with gas, depending on the temperature at the MOX surface. Of the oxygen species, O is commonly chemisorbed at the operating temperature range of 200e400 2 101 C, while O and O2 are mostly contributed at low temperature. The main sensing mechanism of CNT/MOX gas sensors can be described by using the model of a potential barrier to electronic conduction at the grain boundary, as shown in Fig. 6.10. From many studies, it was found that CNTs are embedded in the MOX layer leading to the formation of pen heterojunctions. Therefore, there are two depletion layers to interact with gas, as shown in Fig. 6.10. Two depletion layers are the region on the surface of the MOX and the interface between the CNTs and MOX. The depletion layers at the pen heterojunctions can be modulated. The potential barriers at the interfaces or inside the MOX may be changed. This change of the depletion layer in the pen heterojunctions of CNT/MOX was used to explain the enhanced response of the ﬁlm at low operating temperatures due to the ampliﬁcation effects of junction structure combined with the gas reaction. Moreover, the formation of CNTs in the MOX matrix can also introduce nanochannels. These nanochannels play an important role in gas diffusion. The gas molecules can easily transport into the gas sensing layers, leading to increasing sensitivity.81,82

6.7 Fabrication of electrodes and CNT/polymer nanocomposites for textile-based sensors The fabrication of textile-based sensors is preceded by two processes: (1) preparation of textile-based electrodes and (2) preparation of CNT- polymer nanocomposite materials. 6.7.1 Preparation of textile-based electrode Textile-based electrodes were fabricated using three different techniques, crocheting, embroidery, and screen printing technique, which will be described in the following sections.

6.7.1.1 Crocheting technique This technique was used to make thread-based electrodes. In the ﬁrst step, 100% cotton threads, size #40 thread (lustrous 3-ply) was used as the electrodes.102 The thread-based electrode was designed to look like a Hybrid materials with carbon nanotubes for gas sensing 207

O– – – O– – NO2 – NO2 – NO O2 – NO O2 – NO2 SWCNTs

– SnO2 NO2 ) 2 – O2 – O2 – – NO2 Depletion layer NO2

Ec In oxidizing gas(NO In air d d 1 3 E d f d2 4 Potential

Grain boundary Distance Figure 6.10 Model of a potential barrier to electronic conduction at grain boundary for 77 hybrid single-walled carbon nanotube (SWCNT)/SnO2 sensors. The dashed and continuous lines show the potential barriers of hybrid SWCNT/SnO2 in environment of air and environment of oxidizing gas (i.e., NO2), respectively.

“dumbbell” comprising thread-based sensing materials connected with snap poppers metal electrodes at both ends. To increase the active surface of the sensor, a cotton thread was manually woven into a continuous crochet chain with the length of approximately 10 mm by using crochet hooks. Snap poppers metals were used as electrodes for making electrical contact with an external circuit as shown in Fig. 6.11(a)e(c).

6.7.1.2 Embroidery technique This technique is used to make fabric-based electrodes. In the ﬁrst step, conductive thread was used as the interdigitated electrodes on embroidery on the fabric substrates along the lines to produce comblike patterns as shown in Fig. 6.12 The embroidered interdigitated electrodes were important components of the wearable sensor to function as sensing areas, whose surface area was approximately 0.5 1.5 cm2 in case of fabric-based embroidered electrode (pattern 1). Conductive thread and snap fasteners were used to connect the chemical gas sensors with the external circuit port, in which external devices can be plugged into the active area to fulﬁll the measurement.103 208 Thara Seesaard et al.

Figure 6.11 Fabrication process of thread-based electrodes using crocheting technique. (a) and (b), cotton thread and snap poppers metal were used to make an electrode by manually weaving thread into a continuous crochet chain and (c) the size of complete crochet electrode model has a length of about 10 mm.

Fabric-based embroidered electrode (Pattern 1)

5 mm

15 mm 2 mm

50 mm

S1 S2 S3 S4 Embroidery technique S5 S6 S7 S8 30 mm

1 mm Fabric-based embroidered electrode (Pattern 2) Figure 6.12 Fabrication process of fabric-based embroidered electrodes using embroidery technique.

6.7.1.3 Screen printing technique The screen printing process with conductive silver ink is commonly used in the electronic industry, especially for complicated electronic circuitry and in the electronic components. In the past decade, several groups of researchers conducted experiments printing onto different kinds of substrates such as paper, polyimide, fabric, and glass.104 Thereafter, researchers have Hybrid materials with carbon nanotubes for gas sensing 209

12 3

Fabric substrate Screen printing process

S1 S2 1 2 45° 5 S8 8 3 S3

4 7 4 S7 10 mm S4 5 mm 6 1 mm 5 S6 S5 Screen printed electrode Fabric-based screen printed electrodes Heated 120 °C (15 min) (before heated) Figure 6.13 Fabrication process of fabric-based screen-printed electrodes using screen printing techniques. developed electronic textiles by using screen printing techniques and conductive silver ink to create ﬂexible printed circuits and electronic devices on the textile substrate such as antennas,105 planar electrodes, and planar circuits.106,107 Results of research can describe the methodologies of screen printing for developing a fabric-based screen-printed electrode as fabric-based screen- printed gas sensors for smelling shirt system.108 The screen-printed electrode array fabrication is conducted by the screen printing technique of which processes are shown in Fig. 6.13 (Step 1e4). After completing screen printing with conductive silver ink, the fabric-based screen-printed electrode arrays were oven-dried at 120 C for 15 min (Step 5). Finally, nine metal poppers were attached on the screen-printed electrode arrays to serve as a connector for making electrical connections. 6.7.2 Preparation of CNT/polymer nanocomposite sensing materials The preparation of CNT/polymer nanocomposite gas sensing material for immersion (dip)-coating and drop-coating process are described in detail as shown in Fig. 6.14. Starting with the polymer dissolving process, the ﬁrst step was to dissolve each polymer (3 mg) completely in 1 mL of proper solvent. Then, different types of CNTs such as the MWCNTs, the 210 Thara Seesaard et al.

1 2 3

Polymer solutions Polymer/CNTs solutions Stirred (30 min)

5 4

Polymer/CNTs sensing materials Sonicated (30 min) Figure 6.14 Preparing sensing materials for drop-coating process. carboxylic functionalized single-walled carbon nanotubes (SWCNTs- COOH), and the hydroxyl functionalized single-walled carbon nanotubes (SWCNTs-OH) was blended into each polymer solution to obtain a high conductivity with a percent loading for polymer:CNTs of 70:30 (step 2). Next, this solution was stirred for 30 min, followed by 30 min of continuous ultrasonic vibration at room temperature 25 C. This process (steps 3e4in Fig. 6.14) was repeated three times to ensure sample uniformity.

6.8 Sensor assembly for textile-based gas sensors Textile-based gas sensors were fabricated using two different manufacturing techniques: immersion (dip)-coating technique and drop- coating technique. 6.8.1 Immersion-coating technique For coating the crochet thread with the nanocomposite sensing materials, the solution of CNT/polymer was prepared according to section 11.7.2. Then the “dumbbell” was immersed for 3 h in the solution of coating material under controlled temperature (35 C). Then it was kept at room temperature (about 25 C) under constant stirring for 2 days. Finally, the resulted thread-based gas sensors were baked in the oven at 100 C for 3 h to remove any residual solvents. The immersion (dip-)-coating process and the appearance of the thread-based gas sensor are shown in Fig. 6.15. Hybrid materials with carbon nanotubes for gas sensing 211

Step 1 Step 2 Step 3

Immersion conditions Stirring conditions Bake conditions - Time (3 h) - Time (2 days) - Time (3 h) - Temperature (35 °C) - Temperature (25 °C) - Temperature (100 °C) Thread-based gas sensor Figure 6.15 Immersion-coating technique for thread-based gas sensor fabrication. 6.8.2 Drop-coating technique The CNT/polymer sensing solution mixtures, as prepared by the process described in section 11.7.2, were drop-coated onto the fabric-based embroidered and screen-printed interdigitated electrodes. The electrical resistance of fabric-based gas sensor was measured during the drop-coating process. It was found that the electrical resistance of each sensor was approximately 1e20 kU. Then, the fabricated sensors were baked in an oven at a controlled temperature of 80 C for 1 h to eliminate any remaining solvents as shown in Fig. 6.16. Finally, the shape and features of the fabric-based gas sensors fabricated by both techniques (embroidery and screen printing techniques) are shown in Fig. 6.17.

Fabric-based embroidered electrode Fabric-based embroidered gas sensors

S1 S2 1 2 5 mm 45° S8

8 3 S3 10 mm 10 Heated 80 °C (1 h)

1 mm

7 4 S7 S4

6 5 S6 S5 Fabric-based screen Fabric-based screen printed electrode printed gas sensors Figure 6.16 Drop-coating technique for fabric-based gas sensor fabrication. 212 Thara Seesaard et al.

(a) (b)

Fabric-based embroidered Fabric-based screen printed gas sensors gas sensors

Figure 6.17 Photographs of the fabric-based gas sensors (a) fabricated by embroidery technique and (b) by screen printing technique.

6.9 Characterization of CNT/polymer nanocomposites sensing materials on textile substrate 6.9.1 Scanning electron microscopy 6.9.1.1 Fabric-based embroidered gas sensors SEM was performed to investigate the microstructure of the fabric-based embroidered chemical gas sensors. Fig. 6.18 (a) and (b) show the surface of the conductive thread (functioning as interdigitated electrodes) embroidered on the cotton satin fabric substrate at a magniﬁcation of 30 and 100, respectively.

Figure 6.18 Scanning electron microscopy pictures of (a) the surface of the conductive thread (functioning as interdigitate electrodes) embroidered on the cotton fabric substrate at a magniﬁcation of 30 and (b) the surface of the cotton fabric substrate at a magniﬁcation of 100.63 Hybrid materials with carbon nanotubes for gas sensing 213

Figure 6.19 Scanning electron microscopy pictures of (a) the cross section of the conductive thread and the cotton fabrics as coated by a thick ﬁlm of the CNT/polymer nanocomposites materials 300 and (b) the cross section at 600 magniﬁcation.63

In Fig. 6.19 (a) and (b), the cross section of the conductive thread and the cotton satin fabrics as coated by a thick film of the CNT/polymer nanocomposite materials are displayed at 300 and 600 magnification. It can be seen that the nanocomposite materials have penetrated into the fabric and coated around individual fibers within the thread throughout the full thickness of the cotton satin fabrics. The rough and porous nature of the fabric surface helps to increase the percolation of the analyte gases into the sensing materials, thereby enhancing the sensing response to specificgases.

6.9.1.2 Fabric-based screen-printed gas sensors In Fig. 6.20(a)e(d), the microstructure of the fabric-based screen-printed gas sensors as investigated using SEM is presented: (a) shows at a magnification of 50 the conductive silver ink confirming that the ink was well dispersed on the cotton fabric surface to create the patterned film electrodes; (b) the cross section of the thick film of conductive silver ink covered on the surface of cotton fabric at 150 magnification; (c) the screen-printed electrode surface as coated by a thick film of the CNT/polymer nanocomposite materials at 40 magnification; and (d) the morphology at 600 magnification of the sensing materials infiltrating into the space between the fibers creating good adhesion at the fiberematrix interface. 214 Thara Seesaard et al.

Figure 6.20 Scanning electron microscopy micrographs of the fabric-based screen- printed gas sensors. (a) the conductive silver ink covering the fabric surface without any crack and creating the patterned of electrodes at 50 time magnifications, (b) the cross-section of the thickness around 100 micrometer of conductive silver ink covering on the fabric surface at 150 time magnifications, (c) the cotton fabric and conductive silver ink coated by sensing materials at 40 time magnifications and (d) the morphology of the sensing materials spreading through the electrode surface at 600 time magnifications.

The adhesion force between molecules, between the cotton fabric substrate, and nanosized particles of silver in the conductive ink and even the CNT/polymer nanocomposite networks can be explained by the texture and surface morphology effects on adhesion stabilization in gas sensing ﬁlms deposited by drop-coating process. Hence, the cotton fabric substrate is a porous structure; the conductive silver ink and the Hybrid materials with carbon nanotubes for gas sensing 215

CNT/polymer nanocomposite slurry percolated through the interfiber pores by capillary force. Therefore, the roughness and porous structure on the sensing areas are important to the efficacy of the sensor response to volatile gases. In addition, the chemical properties of CNT/polymer nanocomposite sensing materials and cotton fabric substrate adhere more strongly because of the composition of cotton fabric substrate which is mainly of cellulose. The cellulose fiber and the CNT/polymer nanocomposite networks have increased adhesive interactions between molecule forces by van der Waals forces.109

6.10 Sensing mechanism of CNT/polymer nanocomposites sensing materials on fabric substrate Fig. 6.21 shows the microstructure of the fabric-based gas sensors and the mechanism of sensing in the materials coated on the fiber surface. SEM image at 600 magnification of the gas sensor sheet which was a permeable porous elastic fabric whose surface is covered with a CNT/polymer sensing film is shown. Certain nanocomposite film materials used as sensitive sniffer infiltrated into the space between the fibers and created a good adhesion on the fiberematrix interface. In this case, the sensing capability of the sensor

Polymer swelling due to chemical sorption Gas molecules of gas/liquid molecules

Rf > Ri CNT Polymer

Reference resistance (Ri) Resistance gas exposure (Rf)

Desorption/recovery

Sensor response Xs(t) Polymer/CNTs composite materials coated on fabric fibers Xs(0) Resistance

Baseline Time Reference Odorant Odorant off gas Figure 6.21 Mechanism of carbon nanotube (CNT)/polymer sensing materials coated on the ﬁber surface. 216 Thara Seesaard et al. was not only affected by polymer swelling phenomena but also enhanced by the increasing surface of the fabric. Gas molecules can percolate both into microporous structure of the polymer which was coated on the surface and within the ﬁne structure of the fabric. The sensing mechanism of CNT/polymer gas sensors can be described mainly by the polymer swelling behavior resulting in the change of the electronic pathway on CNTs network.110 Moreover, another sensing principle that may possibly cause the change in the electronic property of the sensors is the electron transferring capability between gas molecules and CNTs.

6.11 Conclusion The unique structure and electronic properties of CNTs provide a tremendous potential for construction of not only CNTs and MOX hybrid materials but also textile sensors in the field of gas sensing applications. Advantages for mixing CNTs in MOXs for gas sensors are the reduction of operating temperature and enhancement of sensitivity and selectivity due to the amplification effects of pen heterojunctions with the gas reaction, formation of nanochannels for gas diffusion, high specificsurface area, and increase of charge carriers on the surface. As a result of these advantages, the hybrid CNT/MOX gas sensor may be used instead of the popular commercial MOX gas sensors (such as TGS gas sensors) in the near future. Moreover, CNT/polymer nanocomposites also selected to use as gas sensing materials in the field of textile sensor and wearable technology as it can operate at room temperature and low energy consumption during operation which is suitable for wearable device. The integration of textiles and electronics for wearable technology has caused the wide variety of smart textile innovations. However, innovative e-textile that exists today is not found function for molecular detection using nanomaterial and nanohybrid materials. Therefore, establishing a sniffing e-textile innovation is the biggest challenge to overcome some restrictions imposed by the basic function of clothing. In addition, it is very important to design and develop sniffing e-textile which has to be more comfortable, flexible, bendable, and washable. Moreover, textile gas sensor innovation can not only be easily adapted to biological function that actually happens but is also fashionable and esthetically acceptable, which is a great opportunity for CNT research and development in terms of the textile gas sensor technology in the future. Hybrid materials with carbon nanotubes for gas sensing 217

Acknowledgments Financial support from the Thailand Research Fund/Mahidol University to T.S. through the Royal Golden Jubilee Ph.D. Program (Grant No. PHD/0178/2554) is acknowledged. T.K. acknowledges Mahidol University and National Nanotechnology Center (NANOTEC). C.W. acknowledges Kasetsart University Research and Development Institute (KURDI). References 1. Granitto PM, Biasioli F, Endrizzi I, Gasperi F. Discriminant models based on sensory evaluations: single assessors versus panel average. Food Qual Prefer 2008;19:589e95. 2. Fitzgerald G, James KJ, MacNamara K, Stack MA. ‘Characterisation of whiskeys using solid-phase microextraction with gas chromatographyemass spectrometry’. J Chromatogr A 2000;896:351e9. 3. Vinaixa M, Vergara A, Duran C, Llobet E, Badia C, Brezmes J, Vilanova X, Correig X. Fast detection of rancidity in potato crisps using e-noses based on mass spectrometry or gas sensors. Sensor Actuator B 2005;106:67e75. 4. Shurmer HV, Gardner JW, Corcoran P. Intelligent vapour discrimination using a composite 12-element sensor array. Sensor Actuator B 1990;1:256e60. 5. Shurmer HV, Gardner JW. Odour discrimination with an electronic nose. Sensor Actuator B 1992;8:1e11. 6. Tikk K, Haugen J-E, Andersen HJ, Aaslyng MD. ‘Monitoring of warmed-over flavour in pork using the electronic nose e correlation to sensory attributes and secondary lipid oxidation products’. Meat Sci 2008;80:1254e63. 7. Santonico M, Pittia P, Pennazza G, Martinelli E, Bernabei M, Paolesse R, D’Amico A, Compagnone D, Natale CD. Study of the aroma of artificially flavoured custards by chemical sensor array fingerprinting. Sensor Actuator B 2008;133:345e51. 8. Ragazzo-Sanchez JA, Chalier P, Chevalier D, Calderon-Santoyo M, Ghommidh C. Identification of different alcoholic beverages by electronic nose coupled to GC. Sensor Actuator B 2008;134:43e8. 9. Yu H, Wang J. Discrimination of LongJing green-tea grade by electronic nose. Sensor Actuator B 2007;122:134e40. 10. Kuske M, Romain AC, Nicolas J. Microbial volatile organic compounds as indicators of fungi. Can an electronic nose detect fungi in indoor environments? Build Environ 2005;40:824e31. 11. Zhang S, Xie C, Zeng D, Zhang Q, Li H, Bi Z. A feature extraction method and a sampling system for fast recognition of flammable liquids with a portable E-nose. Sensor Actuator B 2007;124:437e43. 12. James D, Scott SM, Ali Z, O’Hare WT. Chemical sensors for electronic nose systems. Microchim Acta 2005;149:1e17. 13. Choopun S, Hongsith N, Mangkorntong P, Mangkorntong N. Zinc oxide nanobelts by RF sputtering for ethanol sensor. Physica E 2007;39:53e6. 14. Sriyudthsak M, Supothina S. Humidity-insensitive and low oxygen dependence tungsten oxide gas sensors. Sensor Actuator B 2006;113:265e71. 15. Anukunprasert T, Saiwan C, Traversa E. ‘The development of gas sensor for carbon monoxide monitoring using nanostructure of NbeTiO2’. Sci Technol Adv Mater 2005;6:359e63. 16. Watcharaphalakorn S, Ruangchuay L, Chotpattananont D, Sirivat A, Schwank J. Polyaniline/polyimide blends as gas sensors and electrical conductivity response to CO-N2 mixtures. Poly Inter 2005;54:1126e33. 17. Korotcenkov G. Metal oxides for solid-state gas sensors: what determines our choice? Mater Sci Eng B 2007;139:1e23. 218 Thara Seesaard et al.

18. Korotcenkov G. Gas response control through structural and chemical modification of metal oxide films: state of the art and approaches. Sensor Actuator B 2005;107:209e32. 19. Surnev S, Ramsey MG, Netzer FP. Vanadium oxide surface studies. Prog Surf Sci 2003; 73:117e65. 20. Batzill M, Diebold U. The surface and materials science of tin oxide. Prog Surf Sci 2005; 79:47e154. 21. Brinzari V, Korotcenkov G, Schwank J, Lantto V, Saukko S, Golovanov V. Morpho- logical rank of nano-scale tin dioxide films deposited by spray pyrolysis from SnCl4•5H2O water solution. Thin Solid Films 2002;408:51e8. 22. Korotcenkov G, Cornet A, Rossinyol E, Arbiol J, Brinzari V, Blinov Y. ‘Faceting characterization of SnO2 nanocrystals deposited by spray pyrolysis from SnCl4e5H2O water solution’. Thin Solid Films 2004;471:310e9. 23. Korotcenkov G, Macsanov V, Tolstoy V, Brinzari V, Schwank J, Faglia G. Structural and gas response characterization of nano-size SnO2 films deposited by SILD method. Sensor Actuator B 2003;96:602e9. 24. Szezuka D, Werner J, Oswald S, Behr G, Wetzing K. XPS investigations of surface segregation of doping elements in SnO2. Appl Surf Sci 2001;179:301e6. 25. Kawamura F, Takahashi T, Yasui I, Sunagawa I. Impurity effect on [1 1 1] and [1 1 0] directions of growing SnO2 single crystals in SnO2eCu2O flux system. J Cryst Growth 2001;233:259e68. 26. Judeinstein P, Sanchez C. ‘Hybrid organiceinorganic materials: a land of multi-disciplinarity’. J Mater Chem 1996;6:511e25. 27. Matsubara I, Hosono K, Murayama N, Shin W, Izu N. Organically hybridized SnO2 gas sensors. Sensor Actuator B 2005;108:143e7. 28. Wisitsoraat A, Tuantranont A, Thanachayanont C, Patthanasettakul V, Singjai P. Electron beam evaporated carbon nanotube dispersed SnO2 thin film gas sensor. J Electroceramics 2006;17:45e9. 29. Eder D. Carbon nanotube e inorganic hybrids. Chem Rev 2010;110:1348e85. 30. Wang C, Yin L, Zhang L, Xiang D, Gao R. Metal oxide gas sensors: sensitivity and influencing factors. Sensors 2010;10:2088e106. 31. Schierbaum KD, Weimar U, Gopel€ W, Kowalkowski R. Conductance, work function and catalytic activity of SnO2-based gas sensors. Sensor Actuator B 1991;3:205e14. 32. Zhang W-D, Xu B, Jiang L-C. Functional hybrid materials based on carbon nanotubes and metal oxides. J Mater Chem 2010;20:6383e91. 33. Goldoni A, Petaccia L, Lizzit S, Larciprete R. Sensing gases with carbon nanotubes: a review of the actual situation. J Phys Condens Matter 2010;22:013001. 34. Khanderi J, Hoffmann RC, Gurlo A, Schneider JJ. Synthesis and sensoric response of ZnO decorated carbon nanotubes. J Mater Chem 2009;19:5039e46. 35. Moriguchi I, Hidaka R, Yamada H, Kudo T, Murakami H, Nakashima N. A mesoporous nanocomposite of TiO. and carbon nanotubes as a high-rate Li-intercalation electrode material. Adv Mater 2006;18:69e73. 36. Whitsitt EA, Moore VC, Smalley RE, Barron AR. LPD silica coating of individual single walled carbon nanotubes. J Mater Chem 2005;15:4678e87. 37. Chen Y, Zhu C, Wang T. The enhanced ethanol sensing properties of multi-walled carbon nanotubes/SnO2 core/shell nanostructures. Nanotechnology 2006;17:3012e7. 38. Chen J, Zhang W-D, Ye J-S. Nonenzymatic electrochemical glucose sensor based on MnO2/MWNTs nanocomposite. Electrochem Commun 2008;9:1268e71. 39. Zhao X, Johnston C, Grant PS. A novel hybrid supercapacitor with a carbon nanotube cathode and an iron oxide/carbon nanotube composite anode. J Mater Chem 2009;19: 8755e60. 40. Haisong Q, Jianwen L, Edith M. Smart cellulose fibers coated with carbon nanotube networks. Fibers 2014;2:295e307. Hybrid materials with carbon nanotubes for gas sensing 219

41. Panida L, Enrico S, Natthapol W, Reinhard RB, Teerakiat K. A novel wearable electronic nose for healthcare based on flexible printed chemical sensor array. Sensors 2014; 14:19700e12. 42. Ni Luh WS, Brian Y. Reviewdthe development of gas sensor based on carbon nanotubes. J Electrochem Soc 2016;163:B97e106. 43. Timmer B, Olthuis W, van den Berg A. Ammonia sensors and their applicationsda review. Sensor Actuator B Chem 2005;107:666e77. 44. Mittal M, Kumar A. Carbon nanotube (CNT) gas sensors for emissions from fossil fuel burning. Sensor Actuator B Chem 2014;203:349e62. 45. K€arkkanen A, Avarmaa T, Jaaniso R. “Gas sensing properties of SWCNT and Teflon AF composites,” abstracts of the sixth IEEE conference on sensors. IEEE Sensors 2007: 279e80. Atlanta, 28e31 October 2007. 46. Paabo K, Avarmaa T, Jaaniso R, Kodu M, M€aeorg U, Saar A. Synthesis and gas- sensing properties of phenylhydrazine-functionalised single-wall carbon nanotubes in polymer matrix. Cent Eur J Chem 2013;11:945e52. 47. Lymberis A. Wearable smart systems: from technologies to integrated systems. Conf Proc IEEE Eng Med Biol Soc 2011:3503e6. 48. Chan M, Esteve D, Fourniols JY, Escriba C, Campo E. Smart wearable systems: current status and future challenges. Artif Intell Med 2012;56:137e56. 49. Young-Dong L, Wan-Young C. Wireless sensor network based wearable smart shirt for ubiquitous health and activity monitoring. Sensor Actuator B Chem 2009;140: 390e5. 50. Zhihua Z, Tao L, Guangyi L, Tong L, Yoshio I. Wearable sensor systems for infants. Sensors 2015;15:3721e49. 51. Perrier A, Vuillerme N, Luboz V, Bucki M, Cannard F, Diot B, Colin D, Rin D, Bourg J-P, Payan Y. Smart diabetic socks: embedded device for diabetic foot prevention. IRBM 2014;35:72e6. 52. Lymberis A, Paradiso R. Smart fabrics and interactive textile enabling wearable personal applications: R&D state of the art and future challenges. Conf Proc IEEE Eng Med Biol Soc. 2008:5270e3. 53. Matteo S, Alessandro C. Wearable electronics and smart textiles: a critical review. Sen- sors 2014;14:11957e92. 54. Jing S, Chunghin C, Xiaoming T. Luminous fabric devices for wearable low-level light therapy. Biomed Opt Express 2013;4:2925e37. 55. Cherenack K, Koen VO, Liesbeth VP. Smart photonic textiles begin to weave their magic. Laser Focus World 2012;48:63. 56. Van Os K, Cherenack K. Wearable textile-based phototherapy systems. Stud Health Technol Inform 2013;189:91e5. 57. Zhan L, Liwen J, Mariah DW, Xiangwu Z. Electrodeposited MnOx/carbon nanofiber composites for use as anode materials in rechargeable lithium-ion batteries. J Power Sources 2010;195:5025e31. 58. Hertleer C, Rogier H, Member S, Vallozzi L, Langenhove LV. A textile antenna for off-body communication integrated into protective clothing for firefighters. IEEE Trans Adv Packag 2009;57:919e25. 59. Hertleer C, Rogier H, Vallozzi L, Declercq F. A textile antennas based on high- performance fabrics. In: Proceedings of 2nd European conference on antennas and propagation; Edinburgh, UK; 2007. p. 1e5. 60. Marianna G, Aleksandrs V, Inese P, Ausma V. Screen printed sensor for enuresis alarm system. Mater Sci Text Cloth Technol 2013;3:12e5. 61. Gordon P, Russel T, Steve B, John T. The development of screen printed conductive networks on textiles for biopotential monitoring applications. Sensor Actuator Phys 2014;206:35e41. 220 Thara Seesaard et al.

62. Pandian PS, Mohanavelu K, Safeer KP, Kotresh TM, Shakunthala DT, Parvati G, Padaki VC. Smart Vest: wearable multi-parameter remote physiological monitoring system. Med Eng Phys 2008;30:466e77. 63. Thara S, Panida L, Teerakiat K. Development of fabric-based chemical gas sensors for use as wearable electronic noses. Sensors 2015;15:1885e902. 64. Iijima S. Helical microtubules of graphitic carbon. Nature (London) 1991;354:56e8. 65. Iijima S, Ichihashi T. Single-shell carbon nanotubes of 1-nm diameter. Nature (London) 1993;363:603e5. 66. Bethune DS, Kiang CH, De Vries MS, Gorman G, Savoy R, Vazquez J, Beyers R. Cobalt-catalysed growth of carbon nanotubes with single-atomic-layer walls. Nature (London) 1993;363:605e7. 67. Guo T, Nikolaev P, Thess A, Colbert DT, Smalley RE. Catalytic growth of single- walled nanotubes by laser vaporization. Chem Phys Lett 1995;243:49e54. 68. Thess A, Lee R, Nikolaev P, Dai H, Petit P, Robert J, Xu C, Lee YH, Kim SG, Rinzler AG, Colbert DT, Scuseria GE, Tomanek D, Fischer JE, Smalley RE. Crystal- line ropes of metallic carbon nanotubes. Science 1996;273:483e7. 69. Endo M, Takeuchi K, Igarashi S, Kobori K, Shiraishi M, Kroto HW. The production and structure of pyrolytic carbon nanotubes. J Phys Chem Solids 1993;54:1841e8. 70. Dai H, Rinzler AG, Nikolaev P, Thess A, Colbert DT, Smalley RE. Single-wall nanotubes produced by metal-catalyzed disproportionation of carbon monoxide. Chem Phys Lett 1996;260:471e5. 71. Maeda Y, Kimura S, Kanda M, Hirashima Y, Hasegawa T, Wakahara T, Lian YF, Nakahodo T, Tsuchiya T, Akasaka T, Lu J, Zhang XW, Gao ZX, Yu YP, Nagase S, Kazaoui S, Minami N, Shimizu T, Tokumoto H, Saito RJ. Large-scale separation of metallic and semiconducting single-walled carbon nanotubes. Am Chem Soc 2005;127:10287e90. 72. Zheng M, Jagota A, Semke ED, Diner BA, McLean RS, Lustig SR, Richardson RE, Tassi NG. DNA-assisted dispersion and separation of carbon nanotubes. Nat Mater 2003;2:338e42. 73. Ihara K, Endoh H, Saito T, Nihey F. Separation of metallic and semiconducting single-wall carbon nanotube solution by vertical electric field. J Phys Chem C 2011; 115:22827e32. 74. Arnold MS, Green AA, Hulvat JF, Stupp SI, Hersam MC. Sorting carbon nanotubes by electronic structure using density differentiation. Nat Nanotechnol 2006;1:60e5. 75. Chen FM, Wang B, Chen Y, Li LJ. Toward the extraction of single species of single- walled carbon nanotubes using fluorene-based polymers. Nano Lett 2007;7:3013e7. 76. Cui J, Yang D, Zeng X, Zhou N, Liu H. Recent progress on the structure separation of single-wall carbon nanotubes. Nanotechnology 2017;28:452001e15. 77. Wei BY, Hsu MC, Su PG, Lin HM, Wu RJ, Lai HJ. A novel SnO2 gas sensor doped with carbon nanotubes operating at room temperature. Sensor Actuator B 2004;101: 81e9. 78. Espinosa EH, Ionescu R, Chambon B, Bedis G, Sotter E, Bittencourt C, Felten A, Pireaux JJ, Correig X, Llobet E. Hybrid metal oxide and multiwall carbon nanotube films for low temperature gas sensing. Sensor Actuator B 2007;127:137e42. 79. Wang J, Liu L, Cong SY, Qi JQ, Xu BK. An enrichment method to detect low concentration formaldehyde. Sensor Actuator B 2008;134:1010e5. 80. Hieu NV, Thuya LTB, Chien ND. Highly sensitive thin film NH3 gas sensor operating at room temperature based on SnO2/MWCNTs composite. Sensor Actuator B 2008;129:888e95. 81. Wongchoosuk C, Wisitsoraat A, Tuantranont A, Kerdcharoen T. Portable electronic nose based on carbon nanotube-SnO2 gas sensors and its application for detection of methanol contamination in whiskeys. Sensor Actuator B 2010;147:392e9. Hybrid materials with carbon nanotubes for gas sensing 221

82. Wongchoosuk C, Wisitsoraat A, Phokharatkul D, Tuantranont A, Kerdcharoen T. Multi-walled carbon nanotube-doped tungsten oxide thin films for hydrogen gas sensing. Sensors 2010;10:7705e15. 83. Sedlackova K, Ionescu R, Balazsi C. TEM investigations on CNT added hexagonal WO3 films for sensing applications. Nano 2008;3:223e7. 84. Zhang H, Du N, Chen B, Li D, Yang D. Carbon nanotube-ZnO nanosphere heterostructures: low-temperature chemical reaction synthesis, photoluminescence, and their application for room temperature NH3 gas sensor. Sci Adv Mater 2009;1:13e7. 85. Sanchez M, Guirado R, Rincon ME. ‘Multiwalled carbon nanotubes embedded in solegel derived TiO2 matrices and their use as room temperature gas sensors’. J Mater Sci Mater Electron 2007;18:1131e6. 86. Sanchez M, Rincon ME. ‘Sensor response of solegel multiwalled carbon nanotubes- TiO2 composites deposited by screen-printing and dip-coating techniques’. Sensor Actuator B 2009;140:17e23. 87. Llobet E, Espinosa EH, Sotter E, Ionescu R, Vilanova X, Torres J, Felten A, Pireaux JJ, Ke X, Tendeloo GV, Renaux F, Paint Y, Hecq M, Bittencourt C. ‘Carbon nanotubee TiO2 hybrid filmsfordetectingtracesofO2’. Nanotechnology 2008;19:375501. 88. Li W, Jung H, Hoa ND, Kim D, Hong S-K, Kim H. Nanocomposite of cobalt oxide nanocrystals and single-walled carbon nanotubes for a gas sensor application. Sensor Actuator B 2010;150:160e6. 89. Stern KH. Metallurgical and ceramic protective coatings. Chapman & Hall, Springer Pub- lisher; 1996. ISBN-10: 041-254-4407; ISBN-13: 978-041-254-440-8. 90. Hellstrom SL. Basic models of Spin coating. Lecture note, Stanford University; 2007. 91. 91a Bornside DE, Macosko CW, Scriven LE. Spin coating of a PMMA/chloroben- zene solution. J Electrochem Soc 1991;138:317e20.91b Bornside DE, Macosko CW, Scriven LE. Spin coating: one-dimensional model. J Appl Phys 1989;66:5185. 92. Zhang L, Fang X, Ye C. Controlled growth of nanomaterials. World Scientific Publishing; 2007. ISBN-10: 981-256-7283; ISBN-13: 978-981-256-728-4. 93. Burgelman M. Thin film solar cells by screen printing technology. Proc Workshop Microtech Therm Prob Electron 1998:129e35. 94. Fox IJ, Bohan MFJ, Claypole TC, Gethin DT. Film thickness prediction in halftone screen-printing. J P Mech Eng 2003;217:345e9. 95. Landau LD, Levich BG. Acta physiochim, vol. 17. U R S S; 1942. p. 42e54. 96. Schmidt H, Mennig M. Wet coating technologies for glass. Tutorial: Institut fur€ Neue Materialien; 2000. 97. Mishra VN, Agarwal RP. Effect of electrode material on sensor response. Sensor Actu- ator B 1994;22:121e5. 98. Tamaki J, Miyaji A, Makinodan J, Ogura S, Konishi S. Effect of micro-gap electrode on detection of dilute NO2 using WO3 thin film microsensors. Sensor Actuator B 2005; 108:202e6. 99. Jain U, Harkera AH, Stoneham AM, Williams DE. Effect of electrode geometry on sensor response. Sensor Actuator B 1990;2:111e4. 100. 100a Connolly JR. Introduction quantitative X-ray diffraction methods. Spring 2010.100b Chung FH. Quantitative interpretation of X-ray diffraction patterns. I. Matrix-flushing method of quantitative multicomponent analysis. J Appl Crystallogr 1974;7:519e25. 101. Cheong HW, Lee MJ. Sensing characteristics and surface reaction mechanism of alcohol sensors based on doped SnO2. J Ceram Process Res 2006;7:183e91. 102. Thara S, Panida L, Sasiprapa S, Chayanin K, Teerakiat K. A novel creation of thread- based ammonia gas sensors for wearable wireless security system. In: The 2014 11th international conference on electrical engineering/electronics, computer, telecommunications and information technology. ECTI-CON; 2014. p. 1e4. 222 Thara Seesaard et al.

103. Thara S, Panida L, Teerakiat K. Wearable electronic nose based on embroidered amine sensors on the fabric substrates. In: The 2012 9th international conference on electrical engineering/electronics, computer, telecommunications and information technology. ECTI-CON; 2012. p. 1e4. 104. Saleem K, Leandro L, Ravinder SD. Technologies for printing sensors and electronics over large ﬂexible substrates: a review. IEEE Sens J 2015;15:1e4. 105. Tomas B, Richard L, Jan R. Screen printed antennas on textile substrate. In: Electronics system-integration technology conference (ESTC). vols. 1e4; 2014. 106. Hyejung K, Yongsang K, Binhee K, Hoi-Jun Y. A wearable fabric computer by planar-fashionable circuit board technique. In: Conference: sixth international workshop on wearable and implantable body sensor networks. Berkeley, CA, USA: BSN; 2009. 107. Kai Y, Chris F, Russel T, Steve B, John T. Screen printed fabric electrode array for wearable functional electrical stimulation. Sensor Actuator 2014;213:108e15. 108. Thara S, Panida L, Sasiprapa S, Teerakiat K. On-cloth wearable E-nose for monitoring and discrimination of body odor signature. In: The 2014 IEEE ninth international conference on intelligent sensors, sensor networks and information processing. ISSNIP; 2014. p. 1e5. 109. Randjbaran E, Zahari R, Majid DL, Sultan TH, Mazlan N. ‘Reasons of adding carbon nanotubes into composite systems ereview paper’. Mech Mech Eng 2017;21:549e68. 110. Lorwongtragool P, Wisitsoraat A, Kerdcharoen T. An electronic nose for amine detection based on polymer/SWNT-COOH nanocomposite. J Nanosci Nanotechnol 2011;11:10454e9. CHAPTER SEVEN

Carbon nanomaterials functionalized with macrocyclic compounds for sensing vapors of aromatic VOCs

Pierrick Clément1, Eduard Llobet2 1Microsystems Laboratory, Ecole Polytechnique Féderale de Lausanne (EPFL), Lausanne, Switzerland 2MINOS-EMaS, Department of Electronic Engineering, Universitat Rovira i Virgili, Tarragona, Spain

Contents

7.1 Introduction 223 7.2 Cyclodextrins 226 7.3 Calixarenes and derivatives 229 7.4 Deep cavitands 230 7.5 Conclusions 232 Acknowledgments 235 References 235

7.1 Introduction BTEX is a term describing a set of chemicals closely related to benzene. This set consists of benzene itself, toluene (i.e., methyl benzene), ethylbenzene, and xylenes. BTEX compounds are aromatic volatile organic compounds (VOCs). They are colorless, sweet-smelling liquids under normal temperature and pressure. However, their moderate to high vapor pressures imply that they can evaporate easily. There are natural sources of BTEX compounds. For example, these appear in gas emissions from volcanoes and forest ﬁres, are present in crude oil, and can be found near natural gas and petroleum deposits. However, the main emissions of BTEX into environment are of anthropogenic nature. Primary releases of these compounds occur through emissions from combustion engines, mostly from vehicles and also from aircraft and petroleum coke ovens. Incidentally, BTEX compounds are also found in the smoke of

Semiconductor Gas Sensors, Second Edition ISBN: 978-0-08-102559-8 © 2020 Elsevier Ltd. https://doi.org/10.1016/B978-0-08-102559-8.00007-0 All rights reserved. 223 j 224 Pierrick Clément and Eduard Llobet cigarettes. Petrochemical industry is one of the main emission sources and BTEX compounds are among the most abundantly produced chemicals in the world. For instance, these compounds are created or used during the processing of petroleum products and the manufacturing of many chemical products such as paints, lacquers, thinners, solvents, adhesives, or inks. The manufacturing of rubber and plastics, cosmetics, and pharmaceutical products is also a source of BTEX. Although BTEX can be briefly bound to soils and sediments or be spilled in sea water, most releases eventually end up in the atmosphere (e.g., in land reclamation), where they may react with other pollutants and contribute to the formation of photochemical smog. As any other VOCs, BTEX compounds also play a role in the formation of ground- level ozone which can damage crops and exacerbate respiratory conditions in humans (e.g., asthma). The most common human exposure to BTEX compounds results from contaminated air breathing, particularly in areas of heavy motor vehicle traffic, petrol stations, motor vehicle repair stations, roadside works, and through cigarette smoke. Exposure to BTEX at normal environmental concentrations, and even to higher concentrations over a short period of time, is unlikely to cause significant health damage. However, long-term exposure to higher concentrations (usually only experienced in occupational settings) can be toxic to the liver, kidneys, central nervous system, and eyes. Table 7.1 shows exposure levels for BTEX compounds established by the US safety and health administrations.

Table 7.1 Exposure levels to BTEX compounds as indicated by OSHAa and NIOSHb, January 2018. Compound TWAc (ppm) STELd (ppm) IDLHe (ppm)

Benzene 1 5 500 Toluene 100 150 500 Ethylbenzene 100 125 800 Xylenes 100 150 900 aOSHA: Occupational Safety and Health Administration (USA). bNIOSH: The National Institute for Occupational Safety and Health (USA). cTWA (time-weighted average): Employer shall assure that no employee is exposed to an airborne concentration of the pollutant in excess of the TWA value as an 8-hour TWA. dSTEL (short-term exposure limit): The employer shall assure that no employee is exposed to an airborne concentration of the pollutant in excess of the STEL value as averaged over any 15-min period. eIDLH: Immediately dangerous to life and health exposure level. Carbon nanomaterials functionalized with macrocyclic 225

BTEX compounds feature similar structures but quite different toxico- logical properties. Indeed, benzene is listed among the most harmful VOCs because it is recognized as a human carcinogen by the US Environ- mental Protection Agency and by the European Commission.1,2 Long-term exposures to relatively low concentrations of benzene over months or years lead to severe hemotoxic effects such as aplastic anemia and pancytopenia e and to acute nonlymphocytic leukemia.3 7 According to the Directive 2008/50/EC of the European Parliament and of the Council of May 2008, the limit value for the annual average exposure to benzene is 5 mgm 3 (1.6 ppb).8 Nowadays, several methods for detecting traces of BTEX in air are in use. Most of them involve pumping of the sample and subsequent analysis by employing colorimetric detector tubes or gas chromatography (GC-FID, GC-MS). These methods are bulky, expensive, and do not allow implementation for a continuous monitoring of BTEX traces. In the last few years, preconcentration methods and GC equipment have been improved in terms of miniaturization and with a limit of detection e (LOD) reaching the ppb level for benzene.9 11 However, such systems are still limited by their long response time, high power consumption, and high cost. Alternatively, the use of portable photoionization detector (PID) has been reported as well, but PID devices are not selective and give a total reading for VOCs. The only option to make PID more selective for BTEX in general and benzene in particular is to utilize a single-use, disposable, and rather expensive filter at the inlet port of the device that would result in a dramatic cost increase of running benzene measurements. The industries in which normal activity may result in active exposure of their workers to BTEX compounds (specially to benzene) would clearly benefit from affordable, portable, highly sensitive, and selective detectors able to run continuous measurements. During the last decades, a great effort has been done aiming to investigate different strategies to improve the selectivity of chemical sensors. To reach that milestone, sensors have been equipped with bioreceptors employing specific antigeneantibody-type binding interactions (inspired by nature) where the size, shape, and charge allow the selective detection of biological target species such as proteins, bacteria, viruses, and DNA. Additionally, this new generation of sensors mainly operates in aqueous media where the energy cost for the molecular recognition is reduced because of the water molecules that temporally “occupy” the bioreceptor prior the recognition. For gaseous species, the recognition is different, and 226 Pierrick Clément and Eduard Llobet nonspecific interactions dominate their affinity with the medium. Scienti- fic groups have further exploited the possibility to synthesize molecular receptors that could eventually mimic the specificity of biological receptors reproducing the concept of binding site complementarity and shape recognition. Macrocyclic compounds such as cyclodextrins (CDs), calixarenes, and cavitands have been widely employed because of their common presence of cavities with molecular dimensions, which can act as molecular receptors. In the last years, many research groups have reported the development of sensors employing carbon nanomaterials such as carbon black, carbon nanofibres, carbon nanotubes, and graphene.12 These materials are very attractive because they allow for developing simple, chemoresistive devices operating at low temperatures, even at room temperature. Although carbon nanomaterials are particularly sensitive to their local chemical environment of the gas phase, their functionalization seems essential if the aim is to selectively detect a few target gases or vapors. Indeed, different approaches have been reported for functionalizing carbon nanomaterials in view of tailoring their gas sensing properties. Most of these functionalization strategies consist of creating controlled defects, decorating the outer sidewalls of nanofibres, nanotubes, or the surface of graphene with metal or metal oxide nanoparticles, grafting functional groups such as carbonyl, carboxyl, or amine groups or more complex molecules such as macrocyclic compounds.13 Here we will review the approach of employing macrocyclic compounds grafted to carbon nanomaterials for developing gas sensors and sensor systems, with special emphasis in the results achieved for selectively detecting BTEX compounds. In such an approach, carbon nanomaterials play the role of transducing element (able to collect and transport efficiently electronic charge) and the grafted macrocyclic compounds are the selective molecular receptors, i.e., implement a receptor function in the gas sensor.

7.2 Cyclodextrins CDs are macrocyclic oligosaccharides which contain a hydrophobic cavity presenting hydroxyl groups at both rims that make them water-soluble. The most common CDs consist of six, seven, or eight a-D-glucopyranose units conjoined through a-(1/4)-glycosidic bonds and are, respectively, named a-, b-, g-CDs. They are, therefore, suitable to capture hydrophobic guests in aqueous media, where numerous hosteguest complexes have been reported. Nevertheless, in the solidegas interface, the selectivity is mainly Carbon nanomaterials functionalized with macrocyclic 227 driven by London dispersion interactions, size, and shape fit. Selectivity can be increased by modifying the chemical groups on both rims. Table 7.2 summarizes some applications of modified CDs in gas sensing. Only few examples of carbon nanomaterials functionalized with CDs have been reported so far. Duarte and coworkers18 developed a conductive polymer nanocomposite (CPC) chemoresistor based on linear and branched polyamides synthesized from bifunctional and heptafunctional b-CD mono- mers and (Z) octadec-9-enedioic-N-hydroxysuccinimide ester bearing a multiwalled carbon nanotube (MWCNT) conducting architecture. The latter sensor was formed through a spray deposition of the carbon nanotubes and the CD polymers (dispersed separately in an organic solvent) layer by layer on interdigitated ceramic substrate. The same group has demonstrated the ability of CPC-based gas sensor to reversibly detect polar and nonpolar VOCs with an expected LOD to lay in the low ppb range. Furthermore, polyamide synthesized from b-CD(NH2)2(OH)19 is shown to be selective toward propanol in nitrogen gas carrier. This happens because of the strong hydrophilic character that the 19 hydroxyl moieties offer to the compound making it able to generate many hydrogen bonds with polar protic solvents. Employing the same principle, Nag and coworkers19 developed a quantum resistive chemical vapor sensor based on an array of b-CD functionalized reduced graphene oxide (RGO). Pyrene adamantane was used to noncovalently tether the CDs to the RGO by self-assembly. This innovative connection allows the pep stacking of pyrene with graphene in one end and the inclusion of adamantane in CD cavity in the other end, preserving the accessibility of the analytes to functional sites (route a). They also compared a parallel route (route b) by simply noncovalently binding perbenzylated CD with RGO (RGO@PBCD) (Fig. 7.1). The CD-modified graphene was sprayed layer by layer onto interdigitated electrodes controlling the resistance of the device. The authors demonstrated the selective detection of benzene as low as 400 ppb with a signal-to-noise ratio of 88 with the RGO@PBCD without sensitivity to humidity in nitrogen carrier gas. Following the strategy of employing a sensor array rather than a single sensor, Pi-Guey Su and coworkers20 designed an array of quartz crystal microbalance (QCM) sensors allowing the differentiation of NH3 (1000e5000 ppm), CO (1500e7500 ppm), and NO2 (10e50 ppm) from their tertiary mixture. This discrimination was possible by treating the data of the sensors by principal component analysis. Graphene oxide (GO), b-CD functionalized GO, and N-substituted pyrrole derivative-based films 228

Table 7.2 Examples of modiﬁed cyclodextrins (CDs) with their gas sensing properties. Limit of Modiﬁed CD Transducer Selective to Interferent(s) detection References

2,6-Per-O-(t-butyl- Quartz crystal Benzene Methane, propane, butane, 0.088 mg [14] dimethylsilyl)-a-CD microbalance pentane, eethyne, ammonia, dm 3 in nitrobenzene, air and toluene Polyaniline-b-CD composite Resistive Toluene Benzene N/C [15] Potassium iodide and a-CD Optical Ozone Humidity Several ppb [16] in air g-CD and potassium ions Gas Formaldehyde N/C N/C [17] Llobet Eduard and Clément Pierrick sorption analyzer Carbon nanomaterials functionalized with macrocyclic 229

COOH

COOH OH

O OH HOOC O O OH COOH Pyrene-adamantane OH COOH

N2H4 @ 100°C 24 h reflux Perbenzylated cyclodextrin

RGO@PYAD-CD RGO@PBCD Figure 7.1 Two routes synthesis of functionalized cyclodextrin (CD) to reduced graphene. Reproduced from Nag S, Duarte L, Bertrand E, Celton V, Castro M, Choudh- ary V, Guegan P, Feller JeF. Ultrasensitive QRS made by supramolecular assembly of functionalized cyclodextrins and graphene for the detection of lung cancer VOC biomarkers. J Mater Chem B 2014;2:6571e9. © Royal Society of Chemistry, 2014, with permission. were used as sensitive layers for gas sensing. The b-CD was noncovalently attached to the GO through pep stacking by simply mixing them. Each material was deposited on the QCM by spin coating.

7.3 Calixarenes and derivatives Calixarenes are similar to the CDs’ cyclic structure and have generated interest over the last century because of their easy and tunable synthesis as macrocyclic molecular receptors. They are composed of phenolic units joined at meta position through methylene bridges. Calix[n]arenes are pref- erably synthesized with n ¼ 4, 6, 8 (building blocks) units. Hence, they possess variable cavity dimension with the possibility to functionalize their upper and/or lower rim to tailor their affinity with a target guest molecule through different noncovalent interactions such as pep stacking, cation-p and CH-p interaction, and hydrogen bonding. A review on calixarene-based materials for gas sensing application written by Chilin Zou and coworkers highlights the new development in 230 Pierrick Clément and Eduard Llobet the field of monitoring and detection of hazardous gases.21 Table 7.3 shows some examples of gas sensing applications employing modified calixarenes. Despite the numerous examples that can be found in the literature about the ability of calixarenes for trapping gas molecules with high affinity, only one work on calixarene-functionalized carbon nanomaterials for gas sensing has been reported recently. Baysak and coworkers27 report the use of single-walled carbon nanotubes (SWCNTs), the sidewalls of which were noncovalently functionalized with pyrene bearing calix[4]pyrrole. Sensors were implemented as chemoresistors by coating a filter paper with calixarene-functionalized SWCNTs contacted with two planar electrodes. Fast response and higher affinity for acetone (20e500 ppm) compared with other VOCs were reported. Nevertheless, recent reports, where the calixarene is attached to the carbon nanomaterial via pep e stacking28 30 (including pyrene modification of the calixarene) or covalent bonding31,32 or incorporated in a composite,33 have demonstrated selective recognition of target analytes, but in aqueous media only.

7.4 Deep cavitands Derived from resorcinarene scaffolds, deep cavitands have been widely studied for their synthetic versatility and selective complexation with target molecules. Notably, deep cavitands can be designed by, respectively, tuning their bridging group connected to the phenolic moieties of the resorcinarene. As a result, it is possible to control the dimensions, shape, and binding groups of the formed cavity. Cram and coworkers were the pioneers to study cavitands as potential molecular receptors via the hosteguest strategy.34 Dalcanale and coworkers did a subsequent work by modifying the bridging group of the resorcinarene to monitor VOCs in air. They have recently published a review highlighting their progress.35 Briefly, in their last study, they found out that rigidifying the cavity of the quinoxaline cavitand (QxCav) introducing four ethylenedioxy bridges at the upper rim (EtQxBox) improves the interaction with aromatic guests compared to the conforma- tional mechanism of the QxCav.36 This subtle modification allows additional interaction with toluene, ethylbenzene, and xylene guest than with benzene because of the upper rim that is too far to interact with benzene. They further implemented the EtQxBox as preconcentrator coupled to miniaturized PID, and by using Table 7.3 Modified calixarenes with their gas sensing properties. Modified calixarene Transducer Selective to Interferent(s) Limit of detection References

[22] 5,11,17,23-Tetrakis(tert- Optical NO2/N2O4 N/C N/C butyl)-25-carboxymethoxy-26,27, (colorimetric) 28-tris(ethoxycarbonyl methoxy)calix[4]arene polymer Calix[4]arenes derivatives Quartz crystal N/C Chloroform, N/C [23] microbalance benzene, (QCM) toluene, and ethanol 25,27-(Dipropylmorpholino QCM, surface N/C Dichloromethane, 1.48 ppm for [24] acetamido)-26,28-dihydroxy plasmon chloroform, dichloromethane calix[4]arene resonance benzene, and in air toluene Calix[4]azacrown Luminescence Tetrahydrofuran Acetone, methanol, N/C [25] dichloromethane, ethyl acetate, cyclohexane, n-hexane, benzene, toluene, trifluoroacetic acid, and petroleum ether Calix[4]arenes QCM Methylene Acetone, acetonitrile, 54.1 ppm [26] derivatives chloride carbon tetrachloride, in air chloroform, N,N- dimethylformamide, 1,4-dioxane, ethanol, ethyl acetate, dioxane, xylene, toluene, methanol, n-hexane, and water 232 Pierrick Clément and Eduard Llobet a smart temperature program, benzene is selectively desorbed and its LOD reaching the ppb level. This approach is illustrated in Fig. 7.2. Recently, Llobet and coworkers studied the possibility to couple the promising gas sensing properties of cavitands with MWCNTs as resistive gas sensors.37 They first grafted gold nanoparticles on oxygen plasmaetreated MWCNTs where the thioether-legged QxCav is further tethered on gold by a self-assembled monolayer approach (QxCaveAu-MWCNTs). This functionalization process is illustrated in the upper part of Fig. 7.3. On a sensing event, a charge transfer is observed between the cavitand and the Au-MWCNTs changing the general conductivity of the system. Fig. 7.4 illustrates several response and recovery cycles for the sensors to increasing benzene concentrations in the ppb range. The sensor showed clearly higher sensitivity for benzene than for other aromatic and nonaromatic VOCs, with an LOD of 600 ppt in dry air. Nevertheless, a nonnegligible cross- sensitivity with NO2 and ambient humidity was observed. However, this can be overcome, at least partially, by adding a sensor employing bare Au-MWCNTs, which are extremely sensitive to NO2 and poorly responsive to benzene. The use of a filter at the inlet of the detector would help removing ambient moisture and thus the undesired cross-sensitivity effect of humidity.

7.5 Conclusions The covalent or noncovalent functionalization of carbon nanomaterials with macrocycles opens fascinating opportunities for advancing toward molecular recognition in the gas phase. CDs, calixarenes, and deep cavitands have been employed because they present cavities with molecular dimensions, which can act as molecular receptors. Nowadays, it is possible to finely control the dimensions, shape, and binding groups of the formed cavity. In other words, macrocycles are becoming engineered scaffolds that point toward mimicking the specificity of biological receptors, reproducing the concept of binding site complementarity and shape recognition. In the development of gas sensors employing macrocycles, two main approaches can be identified. On the one hand, macrocycle compounds are employed to functionalize carbon nanomaterials in view of obtaining new functional adsorbent materials, i.e., more efficient and with improved selectivity. These adsorbents are then employed as coatings in gravimetric transducers or in the miniaturized preconcentration units of gas detectors. On the other hand, macrocycle-functionalized carbon nanomaterials are abnnnmtrasfntoaie ihmacrocyclic with functionalized nanomaterials Carbon

(a) (b) (c) Sample in MEMS cartridge 70 PID Referance B = 20 ppb 200 EtQxBox 60 T = 20 ppb cavitand Pump 50 B+T = 20 + 20 ppb R=C H mesh 6 13 150 40

OO O Valve O O O O O 30 N N 100 N N N N N N PID signal (mV) 20 O O O O O O O O MEMS PID 10 50 MEMS cartridge temperature (°C) Filter temperature acquisition 0 control electronic R R 5 6 7 8 9 10 11 12 13 R R Air in EtQxBox electronic Time (min) Figure 7.2 Representation of EtQxBox cavitand structure (a) with a scheme of the benzene monitoring device (b) and typical responses of the photoionization detector (PID) to a temperature ramp (c). Reproduced from Khaled E, Khalil M, el Aziz GA. Calixarene/carbon nanotubes based screen printed sensors for potentiometric determination of gentamicin sulphate in pharmaceutical preparations and spiked surface water samples. Sensor Actuator B Chem 2017;244:876e84. © American Chemical Society, 2013, with permission. 233 234 Pierrick Clément and Eduard Llobet

(a)

O-MWCNT ⊂ Au-MWCNT N2 cav-Au-MWCNT (b)

Figure 7.3 (a) Fabrication scheme of the QxCaveAu-MWCNTs hybrid nanomaterial where Au-RF sputtering is followed by the self-assembly of thioether-legged QxCav monolayer on the Au-NP surface. (b) Schematic representation of the sensing event of a benzene molecule. Adapted from Clément P, Korom S, Struzzi C, Parra EJ, Bittencourt C, Ballester P, Llobet E. Deep cavitand self-assembled on Au NPs-MWCNT as highly sensitive benzene sensing interface. Adv Funct Mater 2015;25:4011e20. © John Wiley and Sons, 2015, with permission.

651.75 40 ppb 651.50 651.25

651.00 20 ppb 650.75 )

Ω 650.50 10 ppb R( 650.25

650.00 5 ppb

649.75 2,5 ppb Air Air Air Air Air 649.50

0 1000 2000 3000 4000 5000 t(s) Figure 7.4 Response and recovery cycles to successively increasing concentrations of benzene for a chemoresistive sensor employing a QxCaveAu multiwalled carbon nanotubes hybrid nanomaterial. Adapted from Clément P, Korom S, Struzzi C, Parra EJ, Bittencourt C, Ballester P, Llobet E. Deep cavitand self-assembled on Au NPs-MWCNT as highly sensitive benzene sensing interface. Adv Funct Mater 2015;25:4011e20. © John Wiley and Sons, 2015, with permission. Carbon nanomaterials functionalized with macrocyclic 235 also used in simple chemiresistive gas sensors in which the carbon nanomaterial is merely an efficient means of conducting free charge carriers and the receptor function is played by the macrocycles attached. The techniques employed in the synthesis of macrocycles and in the functionalization of carbon nanomaterials are well-known and allow for implementing solution processing of gas sensitive devices. This implies that such techniques are suitable for the mass production of both hybrid nanomaterials and sensors at low production costs, allowing cost-effective commercialization. Some of the reported hybrid nanomaterials show remarkable sensitivity and selectivity to aromatic VOCs and, in particular, quinoxaline-bridged cavitand-functionalized MWCNT sensors show very high sensitivity toward low levels of benzene in dry air (i.e., experimentally tested down to 2.5 ppb), with a theoretical lower detection limit of 600 ppt. In addition, both detection and baseline recovery are run at room temperature, which implies that sensors operate at low power consumption. However, sensors still suffer from cross-sensitivity issues, namely to ambient moisture and to oxidizing species such as ozone or nitrogen dioxide. Some solutions exist already to tackle such problems, such as using filters for trapping water at the inlet of the detector system or using an array of sensors with partial selectivity and chemometrics. However, these solutions are suboptimal because filters are cumbersome, may alter the profile of VOCs, or become saturated, and using sensor arrays and pattern recognition adds complexity, makes calibration and recalibration more difficult, and increases overall cost. Should these cross-sensitivity issues be ameliorated by further increasing the performance of functional materials, macrocyclic compound functionalized carbon nanomaterials may soon be integrated in a new generation of inexpensive, handheld analyzers or wearable detectors for BTEX compounds with potential applications in workplace safety or environment monitoring. Acknowledgments E. L. is supported by the Catalan Institution for Research and Advanced Studies (ICREA), via the 2018 Edition of the ICREA Academia Award. References 1. United States Environmental Protection Agency. Drinking water contaminants. 2009. http://water.epa.gov/drink/contaminants/index.cfm#Organic. 2. European Commission. Air quality standards. 2015. http://ec.europa.eu/environment/ air/quality/standards.htm. 236 Pierrick Clément and Eduard Llobet

3. Baan R, Grosse Y, Straif K, Secretan B, El Ghissassi F, Bouvard V, Benbrahim-Tallaa L, Guha N, Freeman C, Galichet L. A review of human carcinogensdpart F: chemical agents and related occupations. Lancet Oncol 2009;10:1143e4. 4. FassO A, Arduino G. Environmental regulation in the European union. Encyclopedia of Environmetrics; 2012. 5. Dougherty D, Garte S, Barchowsky A, Zmuda J, Taioli E. NQO1, MPO, CYP2E1, GSTT1 and GSTM1 polymorphisms and biological effects of benzene exposureda literature review. Toxicol Lett 2008;182:7e17. 6. Hester RE, Harrison RM. Volatile organic compounds in the atmosphere. Royal Society of Chemistry; 1995. 7. DeCaprio AP. The toxicology of hydroquinone-relevance to occupational and environmental exposure. CRC Crit Rev Toxicol 1999;29:283e330. 8. Official journal of the European union,50/EC of the European parliament and of the council of 21 May 2008 on ambient air quality and cleaner air for Europe. 2008. http://ec.europa.eu/ environment/air/quality/legislation/existing_leg.htm. 9. Sun J, Guan F, Cui D, Chen X, Zhang L, Chen J. An improved photoionization detector with a micro gas chromatography column for portable rapid gas chromatography system. Sensor Actuator B Chem 2013;188:513e8. 10. Liaud C, Nguyen N, Nasreddine R, Le Calvé S. Experimental performances study of a transportable GC-PID and two thermo-desorption based methods coupled to FID and MS detection to assess BTEX exposure at sub-ppb level in air. Talanta 2014;127:33e42. 11. Jian R-S, Huang Y-S, Lai S-L, Sung L-Y, Lu C-J. Compact instrumentation of a m-GC for real time analysis of sub-ppb VOC mixtures. Microchem J 2013;108:161e7. 12. Llobet E. Gas sensors using carbon nanomaterials: a review. Sensor Actuator B Chem 2013;179:32e45. 13. Wang F, Swager TM. Diverse chemiresistors based upon covalently modified multiwalled carbon nanotubes. J Am Chem Soc 2011;133:11181e93. 14. Lai CSI, Moody GJ, Thomas JDR, Mulligan DC, Stoddart JF, Zarzycki R. Piezoelec- tric quartz crystal detection of benzene vapour using chemically modified cyclodextrins. J Chem Soc Perkin Trans 1988;2:319e24. 15. Subramanian E, Jeyarani BML, Murugan C, Padiyan DP. Crucial role of undoped/doped state of polyaniline‒b‒cyclodextrin composite materials in determining sensor functionality toward benzene/toluene toxic vapor. 2016. 16. Izumi K, Utiyama M, Maruo YY. A porous glass-based ozone sensing chip impregnated with potassium iodide and a-cyclodextrin. Sensor Actuator B Chem 2017;241:116e22. 17. Wang L, Liang X-Y, Chang Z-Y, Ding L-S, Zhang S, Li B-J. Effective formaldehyde capture by green cyclodextrin based metal-organic framework. ACS Appl Mater Interfaces 2018;10:42e6. 18. Duarte L, Nag S, Castro M, Zaborova E, Ménand M, Sollogoub M, Bennevault V, Feller J-F, Guégan P. Chemical sensors based on new polyamides biobased on (Z) octadec-9-enedioic acid and b-cyclodextrin. Macromol Chem Phys 2016;217: 1620e8. 19. Nag S, Duarte L, Bertrand E, Celton V, Castro M, Choudhary V, Guegan P, Feller J-F. Ultrasensitive QRS made by supramolecular assembly of functionalized cyclodextrins and graphene for the detection of lung cancer VOC biomarkers. J Mater Chem B 2014;2:6571e9. 20. Su P-G, Chuang T-Y. Simple and rapid differentiation of toxic gases using a quartz crystal microbalance sensor array coupled with principal component analysis. Sensor Actuator Phys 2017;263:1e7. 21. Kumar S, Chawla S, Zou MC. Calixarenes based materials for gas sensing applications: a review. J Inclusion Phenom Macrocycl Chem 2017;88:129e58. Carbon nanomaterials functionalized with macrocyclic 237

22. Gusak AS, Ivanova EA, Prokhorova PE, Rusinov GL, Verbitskiy EV, Morzherin YY. Synthesis and use of polymer-immobilized calix[4]arene derivatives as molecular containers for nitrous gases. Russ Chem Bull 2014;63:1395e8. 23. Ozmen M, Ozbek Z, Buyukcelebi S, Bayrakci M, Ertul S, Ersoz M, Capan R. Fabrication of LangmuireBlodgett thin films of calix[4]arenes and their gas sensing properties: investigation of upper rim para substituent effect. Sensor Actuator B Chem 2014;190:502e11. 24. Acikbas Y, Bozkurt S, Halay E, Capan R, Guloglu ML, Sirit A, Erdogan M. Fabrication and characterization of calix[4] arene LangmuireBlodgett thin film for gas sensing applications. J Inclusion Phenom Macrocycl Chem 2017;89:77e84. 25. Oueslati I, Paixao JA, Shkurenko A, Suwinska K, Seixas de Melo JS, Batista de Carvalho LAE. Highly ordered luminescent calix[4]azacrown films showing an emission response selective to volatile tetrahydrofuran. J Mater Chem C 2014;2: 9012e20. 26. Temel F, Tabakci M. Calix[4]arene coated QCM sensors for detection of VOC emissions: methylene chloride sensing studies. Talanta 2016;153:221e7. 27. Baysak E, Yuvayapan S, Aydogan A, Hizal G. Calix[4]pyrrole-decorated carbon nanotubes on paper for sensing acetone vapor. Sensor Actuator B Chem 2018;258: 484e91. 28. Sun Y, Mao X, Luo L, Tian D, Li H. Calix [4] arene triazole-linked pyrene: click synthesis, assembly on graphene oxide, and highly sensitive carbaryl sensing in serum. Org Biomol Chem 2015;13:9294e9. 29. Yang L, Ran X, Cai L, Li Y, Zhao H, Li C-P. Calix [8] arene functionalized single-walled carbon nanohorns for dual-signalling electrochemical sensing of aconitine based on competitive host-guest recognition. Biosens Bioelectron 2016;83:347e52. 30. Yang L, Xie X, Cai L, Ran X, Li Y, Yin T, Zhao H, Li C-P. p-sulfonated calix [8] arene functionalized graphene as a “turn on” fluorescent sensing platform for aconitine determination. Biosens Bioelectron 2016;82:146e54. 31. Dionisio M, Schnorr JM, Michaelis VK, Griffin RG, Swager TM, Dalcanale E. Cavitand-functionalized SWCNTs for N-methylammonium detection. J Am Chem Soc 2012;134:6540e3. 32. Adarakatti PS, Malingappa P. Amino-calixarene-modified graphitic carbon as a novel electrochemical interface for simultaneous measurement of lead and cadmium ions at picomolar level. J Solid State Electrochem 2016;20:3349e58. 33. Khaled E, Khalil M, el Aziz GA. Calixarene/carbon nanotubes based screen printed sensors for potentiometric determination of gentamicin sulphate in pharmaceutical preparations and spiked surface water samples. Sensor Actuator B Chem 2017;244: 876e84. 34. Cram DJ, Karbach S, Kim HE, Knobler CB, Maverick EF, Ericson JL, Helgeson RC. Host-guest complexation. 46. Cavitands as open molecular vessels form solvates. JAm Chem Soc 1988;110:2229e37. 35. Pinalli R, Pedrini A, Dalcanale E. Environmental gas sensing with cavitands. Chem Eur J 2018;24:1010e9. 36. Trzcinski JW, Pinalli R, Riboni N, Pedrini A, Bianchi F, Zampolli S, Elmi I, Massera C, Ugozzoli F, Dalcanale E. In search of the ultimate benzene sensor: the EtQxBox solution. ACS Sens 2017;2:590e8. 37. Clément P, Korom S, Struzzi C, Parra EJ, Bittencourt C, Ballester P, Llobet E. Deep cavitand self-assembled on Au NPs-MWCNT as highly sensitive benzene sensing interface. Adv Funct Mater 2015;25:4011e20. This page intentionally left blank CHAPTER EIGHT

Luminescence probing of surface adsorption processes using InGaN/GaN nanowire heterostructure arrays

Konrad Maier1, Andreas Helwig1, Gerhard Muller€ 2, Martin Eickhoff3 1Airbus Group Innovations, Munich, Germany 2Department of Applied Sciences and Mechatronics, Munich University of Applied Sciences, Munich, Germany 3Institute of Solid State Physics, University of Bremen, Bremen, Germany

Contents

8.1 Adsorptiondkey to understanding semiconductor gas sensors 239 8.2 III-nitrides as an emerging semiconductor technology 243 8.3 Photoluminescent InGaN/GaN nanowire arrays 243 8.4 Optical probing of adsorption processes 245 8.5 Experimental observations of PL response 246 8.5.1 General response behavior 246 8.5.2 Response to oxidizing gases 246

8.5.3 Response to H2O vapor 248 8.5.4 Response to reducing gases 249 8.6 Analysis of adsorption phenomena 250 8.6.1 Concentration and temperature dependence of the PL response 250 8.6.2 Competitive adsorption of air constituents 253

8.6.3 Competition between quenching and enhancing H2O adsorbates 259 8.7 Molecular mechanism of adsorption 261 8.8 Conclusions and outlook 266 References 267

8.1 Adsorptiondkey to understanding semiconductor gas sensors Gas sensors based on metal oxides (MOXs) are a very widely studied e class of sensors.1 4 The most commonly employed transduction mechanism is the detection of electrical resistance modulations of MOX materials as

Semiconductor Gas Sensors, Second Edition ISBN: 978-0-08-102559-8 © 2020 Elsevier Ltd. https://doi.org/10.1016/B978-0-08-102559-8.00008-2 All rights reserved. 239 j 240 Konrad Maier et al. these are exposed to reactive gases while being heated to elevated temperatures. Transduction mechanisms which convert a partial pressure of reactive gas, pgas, in the ambient air into an electrical output signal are twofold: the ﬁrst effect is reductions in surface conductance that take place as strongly oxidizing, i.e., electron withdrawing gases, adsorb on n-type oxides. Widely studied examples are O2,NO2, and O3. The second important effect are en- hancements in surface conductance that take place as reducing gases adsorb on heated n-type oxides and as these interact with coadsorbed oxygen ion species. In these surface reactions, neutral combustion products such as H2O and CO2 are generated and electrons, initially trapped on oxygen ion species, are returned to the semiconductor adsorbent. Examples of reducing gases that follow this second line of detection are CO, H2, and a huge range of hydrocarbons. Experimental parameters that can be derived from resistance measurement and gas exposure tests are the relative resistance response Rres(pgas) and the gas sensitivity Sres(pgas): R0 Rgas Rres pgas ¼ ; (8.1) R0 dRres Sres pgas ¼ . (8.2) dpgas

In these equations, pgas is the partial pressure of the reactive gas in the air ambient, Rres the magnitude of the relative resistance response to the reactive gas, and Sres the corresponding gas sensitivity. R0, in turn, is the baseline resistance of the sensor under clean-air conditions and DRgas ¼ R0 Rgas the negative change in the sensor resistance under gas exposure. Although these processes are qualitatively well understood, a quantitative analysis of those microscopic mechanisms that ultimately lead to the experimentally determined functions Rres(pgas) and Sres(pgas) has remained difﬁcult and continues to be challenging as the electrical conductivity can depend in complex and manifold ways on the material characteristics of the sensitive layers. In the analysis of these functions, it has been customary to break down the experimentally observed sensitivity Sres(pgas) into a transducer e and a receptor part5 7: dR dR dN S p ¼ res ¼ res Ominus res gas dp dN dp gas Ominus gas ¼ TrðRres; NOminusÞRec NOminus; pgas ; (8.3) Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 241

where NOminus stands for the areal density of adsorbed oxygen ion species on the sensor surface. Whereas the receptor part, Rec(NOminus, pgas), is normally considered to follow simple mass-action laws or common adsorption isotherms, the transducer part, Tr(Rres, NOminus), can take on different forms depending on the crystal size, the stoichiometry, and the morphology of the e sensing layers.5 7 Depending on the height and widths of the intergrain potential barriers, electron transport across grain boundaries can proceed through thermal activation and/or tunneling steps. Furthermore, as crystal sizes often have dimensions in the range of nanometers, different types of carrier depletion can abound in the bulk crystal grains, ranging from e regionally depleted to critically and volume-depleted.5 7 In nanocrystalline and porous layers, ﬁnally, the electrical transport over macroscopic distances can often take the form of percolation paths, which makes the interpretation of resistive response data even more complex.8,9 In recent years, another form of chemical response has been studied, e which can be observed in MOX materials grown in nanowire form.10 12 This kind of response involves luminescence light induced by UV light sources and detection of longer-wavelength visible light. The interesting aspect is that reactive gases that adsorb at the nanowire surfaces may act as recombination centers which reduce the luminescence intensity below its clean-air baseline level. As this optically induced chemical sensitivity does not rely on the presence of thermally activated charge carriers, the optical response can in general be observed at more moderate temperatures than the resistive response of MOX materials. Even more important is that the generation and emission of luminescence light is a local phenomenon that does not depend in a similarly complex way on the morphology and intergrain transport of photogenerated charge carriers as in resistively readout sensors. Measurements of the optical response of MOX semiconductors therefore hold promise to shed more light on those processes that are related to the adsorption of gases on MOX surfaces. In this chapter, we report on gas detection experiments performed on arrays of ternary group III-nitride nanowires, namely InGaN nanowires formed on GaN nanowire templates. Such InGaN/GaN heterostructure nanowire arrays (NWAs) are attractive for luminescence studies as the bandgap of InGaN alloys can be varied over a large range extending from Eg z 0.7 eV (InN) up to Eg z 3.5 eV (GaN) with the bandgap always remaining direct.13 Because of this bandgap variability, InGaN/GaN NWAs can be grown which allow the PL excitation and the PL emission light to be absorbed and emitted in well-separated spectral ranges. This spectral 242 Konrad Maier et al. separation allows comparatively simple equipment to be used which is of concern considering future gas sensing applications.14 Other aspects which make InGaN alloys interesting for fundamental investigations into gas sensing mechanisms are that InGaN materials retain their luminescence up to temperatures extending well into the operation temperature range of resistively readout MOX gas sensors. Most importantly, increasing evidence emerges e that InGaN surfaces carry native oxides,15 21 which make InGaN/GaN nanowires behave similar as more conventional MOX nanowires. On the pages below, we will show that the PL response of InGaN/GaN NWAs, RPL(pgas, T), quite universally takes the form of products consisting of a temperature-dependent recombination term asat(T) and a Langmuir adsorption isotherm, qL(pgas, T): PLgas PL0 RPL pgas; Eads; T ¼ ¼ asatðTÞ qL pgas; Eads; T . (8.4) PL0 Whereas the ﬁrst term depends on the physical parameters that control the radiative recombination processes in the volume of the InGaN/GaN nanowires and of the nonradiative ones at their surfaces, the second term depends on those parameters that control the chemistry of the adsorption processes at the transducer surface. These are the adsorbate partial pressure in the ambient air, pgas, the binding energy of the adsorbates on the transducer surface, Eads, and the temperature T at which the adsorption takes place. In this chapter, our concern is on the function qL(pgas, Eads, T) and how it varies under different conditions of gas exposure and sensor operation conditions. Making use of the excellent luminescence properties of InGaN/GaN NWAs, we have been able to study the adsorption behavior of several classes of gases on InGaN surfaces with native surface oxides over wide concentration and temperature ranges. In this way, we have obtained information on adsorbates and adsorbate-binding energies, which hitherto could not be observed and discussed with clarity on other kinds of e gas-sensitive materials.14,22 27 Examples discussed here include the competitive adsorption of oxidizing air constituents on III-nitride surfaces and the surprisingly complex adsorption behavior of H2O, which can form both quenching and enhancing adsorbates on such surfaces, depending on experimental conditions. We further provide evidence that enhancing water adsorbates produced in surface oxidation reactions provide an indirect approach to detect otherwise unreactive hydrocarbon species. Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 243

8.2 III-nitrides as an emerging semiconductor technology With the silicon semiconductor technology approaching its physical limitations, the interest of the semiconductor community has turned to wide-bandgap materials such as silicon carbide (SiC),28 diamond,29 and gallium nitride (GaN).30 Among these materials, GaN is outstanding because of its ability of forming alloys with indium (In) and aluminum (Al). Because of this capability, III-nitride materials form a continuous series of alloys with direct bandgaps ranging from 0.7 up to 6.2 eV.13 With this potential at hand, UV-LEDs and lasers have become available and III-nitride materials have rapidly developed into a reliable materials base for the rapidly developing field of solid-state lighting technology.31 Another attractive feature of the III-nitride alloy system is the spontaneous and piezoelectric polarization phenomena that can be observed in AlGaN/GaN heterostructures.32 These latter phenomena have been important for the realization of high-electron mobility transistors (HEMTs) which have become important building blocks in the fields of high-temperature and high-frequency electronics.33 In addition to these mainstream applications, AlGaN/GaN HEMTs have also received increasing attention in the field of chemical and biochemical sensors. Recent reviews of this work can be found in Ref. 34,35.

8.3 Photoluminescent InGaN/GaN nanowire arrays While so far most of the work on III-nitride-based chemical sensors has used conventional chemical-to-electrical transducer principles, very little work has been reported yet which employs the excellent optoelectronic properties of the III-nitride material system. This chapter attempts to close this gap, reporting on experiments performed on III-nitride materials grown in the form of NWAs on silicon substrates. Whereas our previous work has concentrated on the growth, structural, and physical characterization of such materials,23 our concern here is presenting an in-depth study of those photoluminescence (PL) changes that take place as NWAs are illuminated by low-cost ultraviolet light sources while being simultaneously exposed to e different chemical environments.14,22 27 Fig. 8.1 shows the investigated nanowire heterostructures, summarizing at the same time information about their geometrical size, their crystallographic orientation, and their luminescence properties. The nanowires shown in Fig. 8.1(a) were grown on (111) silicon substrates using plasma-assisted 244 Konrad Maier et al.

(a) (b) Side view Top view GaN top (20 nm) InGaN layer

- (200 nm) (0001) GaN base (500 nm)

Si (111) substrate 300 nm

100 100

10 TNW 10

1 375 K 1 PL intensity (a.u.) PL λ =325 nm PL maximum (a.u.) PL ex

0.1 0.1 2.0 2.2 2.4 2.6 2.8 3.0 3.2 10 100 1000 Energy (eV) T (K) Figure 8.1 (a) Heterojunction nanowires on Si substrates with axial stacking of GaN and InGaN sections, (b) cross-sectional and top view scanning electron microscopy (SEM) pictures of nanowire arrays grown using plasma-assisted molecular beam epitaxy; (c, d) spectral distribution of photoluminescence (PL) light and temperature dependence of PL intensity. molecular beam epitaxy with the growth proceeding along the (0001) axis. Substrate temperatures during the growth of GaN sections were 720C and 500C during the InGaN sections. The nanowires typically consist of a relatively long GaN base (w500 nm), followed by a shorter InGaN section (w200 nm) and a thin GaN cap layer (w20 nm). The scanning electron microscopy (SEM) image in Fig. 8.1(b) shows that the self-assembled growth mode leads to irregular arrays of nanowires with hexagonal cross sections. Gas access to the lateral side walls is enabled by the relatively large wire-to-wire distances which are in the same order of magnitude as the diameters of the nanowires. The cross-sectional SEM shows that the thickness of the nanowires tends to increase toward the growth surface, which is partly caused by the lateral overgrowth of the previously deposited wire sections. The lower-bandgap InGaN sections therefore are likely to carry a thin GaN coating, which further tends to become covered by a thin layer of native e oxide as the NWAs are exposed to the air ambient.15 21 Fig. 8.1(c) shows spectra of luminescence light as emitted in response to excitation light from Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 245 a He-Cd laser at 325 nm and as observed at different NWA temperatures. Fig. 8.1(d), ﬁnally, shows the variation of PL intensity with temperature in the range from 4K up to 375K.

8.4 Optical probing of adsorption processes The measurement system for the characterization of the optochemical response is displayed in Fig. 8.2(a) and the spectral characteristics of its optical components in Fig. 8.2(b). The light of a near-UV LED light source (l w 365 nm), used for luminescence excitation, is reflected onto the InGaN/GaN NWA placed on the ground plate and focused onto the NWA by a lens system. The same lens system captures the luminescence light (green arrow in Fig. 8.2(a)) and focusses it onto the detector window of a compact photomultiplier tube integrated into the top lid of the sensor system. Appropriate filters were used to separate the UV excitation and the green luminescence light. To allow measurements at different transducer temperatures, the NWA samples were mounted on a ceramic heater substrate carrying a screen-printed platinum meander on its backside. This Pt heater meander simultaneously served as a Pt thermometer and as an indicator for the adjusted heater temperature. With this sample holder, maximum temperatures in the range of 150 C could be reached. The total volume of the sensor chamber amounted to approximately 5 cm3. With the total gas flow rate of 500 sccm, this allowed for gas exchange times in the order of one second.

Figure 8.2 (a) Experimental arrangement for measuring the photoluminescence emission spectra of GaN/InGaN nanowire transducers under variable gas ﬂows and at different nanowire array operation temperatures; (b) spectral characteristic of the optical components indicated in (a). PMT, photomultiplier tube. 246 Konrad Maier et al.

Figure 8.3 Schematics of the gas test rig, featuring test gas cylinders and a vapor saturation bottle for producing high concentrations of alcohols, either diluted in SA or in N2. The PL response tests toward different gases were carried out using a custom-designed gas test rig with a set of mass flow controllers as shown in Fig. 8.3. All gas/vapor mixtures delivered from this test gas rig were guided through a downstream mass flow controller into the measurement chamber to maintain a constant gas flow rate of 500 sccm independent of gas composition. In this way, any effects potentially emerging from variable air flows were ruled out.

8.5 Experimental observations of PL response 8.5.1 General response behavior InGaN/GaN nanooptical probesdlike MOXsdexhibit a nonselective broad-range gas response. When exposing InGaN/GaN NWAs to different gases, three different kinds of PL response behaviors can be observed. In Fig. 8.4, these responses are schematically represented as functions of time arising from rectangular boxlike gas exposure proﬁles. In addition, this ﬁgure shows how gas response values were evaluated from such transients. 8.5.2 Response to oxidizing gases

Among all reactive gases, O2 is the one that occurs in highest concentrations (w20% in N2) in ambient air. As a reference point to all other kinds of gas response, it is therefore necessary to investigate how InGaN/GaN nanooptical probes respond to changes in gas concentration from pure N2 to synthetic air (SA: 20% O2 and 80% N2) and back again. Fig. 8.5 shows that such changes produce quenching PL responses, showing that O2 adsorbates form epne o30pbO ppb 330 to responses ya30pbNO ppb 300 a by ocnrto f2%O 20% of concentration iue8.4 Figure nanowire InGaN/GaN using processes adsorption surface of probing Luminescence iue8.5 Figure response. enhancing enhancing to (b) quenching response, from quenching transition pure (c) (a) and transducers: response, array nanowire of transients R

a (b) G (a) T N a htlmnsec P)rsos oa O an to response (PL) Photoluminescence (a) xrcino a epnedt rmpoouiecne(L response (PL) photoluminescence from data response gas of Extraction (a) 2 us nasnhtcarbcgon 2%O (20% background air synthetic a in pulse 2 3 P /N sapidi akrudo r ytei air. synthetic dry of background a in applied as – T 2 sapidi nietbcgon f10 r N dry 100% of background inert an in applied as ( ) b (c) (b) P – G 2 ocnrto us iha with pulse concentration Q T – – 2 8%N /80% ( ) – T 2 2 followed .()PL (b) ). 247 248 Konrad Maier et al. surface recombination centers which enhance the surface recombination velocity beyond its native level in inert N2.AsO2 is a strongly oxidizing gas with ﬁ ¼ 36,37 a positive electron af nity of EO2 0.448 eV, adsorbed O2 molecules are likely to trap electrons photogenerated in the InGaN/GaN nanowires, thus forming negatively charged adsorbates. Such negatively charged centers cause an upward band bending and thus attract photogenerated holes into the electron-trapping adsorbates. As the energy released during such surface recombination processes can be dissipated either inside the nanowires themselves or transferred as kinetic energy to the desorbing molecules, these processes are likely to be nonradiative. NO2 and O3 are even more strongly oxidizing gases with electron afﬁn- ¼ ¼ 36,37 ities of ENO2 2.273 eV and EO3 2.103 eV. Because of their high reactivity, both gases are harmful to human health. Normal environmental concentrations of NO2 are below 1 ppm and below 100 ppb in the case of O3. Being strongly oxidizing gases, both compete with the much higher concentrations of O2 for photogenerated electrons from the NWA bulk and for adsorption sites on the InGaN surfaces. As shown on the right- hand side of Fig. 8.5, sub-ppm concentrations of NO2 and O3 produce reductions in the PL intensity additionally those reductions already induced by the background O2. Quite interestingly, the NO2- and O3-induced reductions are of similar size as the ones produced by the huge background 5 concentration of 20% O2 ¼ 2 10 ppm O2. We will see in the following that NO2 and O3 have a capability of displacing existing O2 adsorbates from their binding sites, thus forming adsorbate sites with strongly enhanced recombination velocities.

8.5.3 Response to H2O vapor

H2O was found to be outstanding among a large variety of gases as it produces PL intensity changes in inert backgrounds of dry N2 without any intervention of reactive O2. Moreover, H2O proved to be exceptional as it can form both quenching (Q) and enhancing (E) adsorbates, which convert into each other 26 during a single vapor exposure. This capability of H2O of forming Q- and E-adsorbates is shown in Fig. 8.6.Inthisﬁgure, PL responses are shown as H2O vapor pulses with 30% relative humidity were applied under different NWA operation conditions. In the left-hand panels, a low LED excitation light intensity of 7 mW was applied, while vapor sensing tests were performed at room temperature and at 120 C. In the right-hand panels, results are shown in which the same experiment was repeated, but at a higher LED intensity of 200 mW. An overall look at these data shows that on onset of Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 249

(a)L (b) H

P Opt = 0.7 mW

– R – P Opt = 200 mW –

T ( ) T ( ) Figure 8.6 Photoluminescence response to water vapor pulses (gray boxes) as applied at different nanowire array temperatures and at different LED light powers used for excitation. each H2O vapor pulse, the PL is initially sharply quenched and that this quenching fades away with increasing speed as the NWA temperature and/ or the LED light intensity are raised. While in the limits of low temperature and low LED power almost purely quenching responses are observed, almost purely enhancing responses are found in the limit of high temperature (120C) and high LED light intensity. Another interesting feature occurs immediately after termination of the H2O exposure pulses. There, PL overshoots are observed which fade away with increasing speed, again as NWA temperatures and/or LED light powers are raised. 8.5.4 Response to reducing gases

In contrast to O2 and H2O, reducing gases such as H2 and hydrocarbons do not have any intrinsic capability of producing sizable PL changes.25,27 In Fig. 8.7, an InGaN/GaN NWA was exposed to increasing concentrations of ethanol (EtOH), either applied in a background of inert N2 or in a more reactive background of synthetic air. Additionally, room temperature and elevated temperature (T ¼ 120 C) exposures are compared. The very small quenching responses toward EtOH, when applied in inert N2 backgrounds, show that EtOH adsorbatesdlike the oxidizing gases discussed abovedintrinsically form recombination centers. Considering the small responses of DPL 1% at concentrations in the order of 104 ppm, the EtOH-derived recombination centers are orders of magnitude less efﬁcient 250 Konrad Maier et al.

(a) (b)

(e) (e)

T ( ) T ( ) Figure 8.7 Photoluminescence response of an InGaN/GaN nanowire array toward ethanol exposure pulses applied in backgrounds of N2 (a, c) and in synthetic air (b, d). Panels (e, f) show the timing of the ethanol vapor ﬂows. than those derived from O2 and even more from NO2 or O3.WhenEtOHis applied in reactive backgrounds of synthetic air, however, EtOH consistently shows sizable enhancing responses, particularly at elevated temperatures.27 Similar data as for ethanol were also obtained for other alcohols and for aliphatic hydrocarbons, not carrying any functional groups.27 Like EtOH, enhancing PL responses of any sizable magnitude consistently were only observed when the reducing gases were diluted in synthetic air and when the InGaN/GaN NWAs were operated at elevated temperatures. Overall, these latter data indicate that reducing gases are indirectly detected via the consumption of quenching oxygen adsorbates and via the formation of enhancing water adsorbates. CO2 molecules, which are also likely to form during surface oxidation processes, did not produce any sizable PL response.27

8.6 Analysis of adsorption phenomena 8.6.1 Concentration and temperature dependence of the PL response Having established the main qualitative features of the gas response, we now turn to the discussion how this response varies with the gas concentration and the temperature of the InGaN/GaN NWA transducers. Fig. 8.8 Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 251

(d) – (a) – – – R – R (e) (b) –

– R R – (f) (c) –

– R – R – – – – C C Figure 8.8 Concentration and temperature dependencies of the photoluminescence

(PL) response of an InGaN/GaN nanowire array to PL-reducing (O2,NO2, and O3) and PL-enhancing gases (H2O and EtOH). Panel f shows the variation of the PL response with LED light intensity. Data points represent measured data; full lines represent fits to the Langmuir adsorption and recombination model.25 summarizes response data for some of the major air constituents and air contaminants (O2,NO2, and O3) as well as for H2O and EtOH. To ensure comparable conditions, all data were acquired at high LED excitation light intensities (w200 mW). The test gases O2 and H2O were diluted in dry N2 to show that these intrinsically form recombination-enhancing (O2) and recombination-reducing (H2O) adsorbates without any interference with other reactive gases. This, of course, does not exclude the possibility that O2 and H2O molecules might be competing with the much more numerous N2 molecules for adsorption sites on the InGaN/GaN surfaces. All other gases were diluted in dry synthetic air, which means that the main competition of these test gases is with the O2 molecules in the synthetic air. Whereas data points stand for measured response data, the full lines represent fits to Langmuir adsorption isotherms (Eq. 8.5). As a common trend, one can observe that all gas responses increase as the InGaN/GaN operation temperatures are raised. The observed magnitudes 252 Konrad Maier et al. of response, however, differ in sign with all oxidizing gases producing negative, i.e., quenching, and H2O and EtOH positive, i.e., enhancing PL responses. Qualitatively, all gases show very weak concentration dependencies with trends toward saturation at very high gas concentrations. The full lines through the individual data sets show that all data can be fitted to e Langmuir isotherms38 40 by choosing optimum values for the adsorption energy Eads and for the saturated gas response asat: 2 3 6 7 6 pgas 7 RPL pgas; T; M ¼ asatðTÞ4 5: Eads pgas þ P00ðT; MÞexp kBT (8.5a)

In the square bracket term, pgas stands for the analyte partial pressure and T for the InGaN/GaN NWA temperature and P00 for the Langmuir desorption pressure of the analyte gas:

kB T P00ðT; MÞ ¼ . (8.5b) VQðT; MÞ

VQ(T, M), ﬁnally, is the quantum volume of the adsorbates:

3 h2 2 VQðT; MÞ¼ ; (8.5c) 2 p MkB T with M standing for the adsorbate molecular mass and h and kB for Planck’s and Boltzmann’s constants. For the gases considered in this chapter, P00 takes 11 24,38 on values in the order of P00 z 10 Pa Although all data can be fitted to such Langmuir isotherms, experimental verification is not easy to perform as the predicted concentration dependencies are very weak, varying in a quasilogarithmic manner around the centers of the sensitivity windows, i.e., around those partial pressures p1/2 at which the InGaN/GaN NWAs attain half of their saturation responses. Experimental verification requires very large variations in the test gas concentrations, which for physical and technical reasons cannot be easily produced in all cases. The best evidence for the Langmuir hypothesis could be obtained in the case of EtOH, where the test gas concentration could be varied over five orders of magnitude (Fig. 8.8(e)). Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 253

–

Figure 8.9 Correlation between adsorption energy Eads and center concentration p1/2 of the sensitivity windows. The full lines show p1/2 versus Eads relationships as calculated 25 from the Langmuir adsorption and recombination model for O2 and NO2 (O3), respectively. The data points stand for pairs of p1/2-Eads values that had been extracted from the experimental data.

Reference to Fig. 8.8 shows that the positions of the sensitivity windows, i.e., those partial pressures p1/2 at which 50% of the saturation PL response, is observed, depends on the respective species. Equating the right-hand side of Eq. (8.5a) to RPL(p, T) ¼ asat(T)/2, the partial pressures p1/2 can be evaluated at which the gas response attains half of its maximum possible response: ¼ Eads : p1=2 P00exp (8.6) kBT In Fig. 8.9, this quantity is plotted as a function of the Langmuir adsorption and recombination (LAR) adsorption energy Eads. Comparison of the theoretical results (full lines) with the values of Eads (data points) extracted from the data of Fig. 8.8 shows that relatively modest variations in adsorption energy lead to considerable shifts in the positions of the sensitivity windows. 8.6.2 Competitive adsorption of air constituents Focusing on the positions of the sensitivity windows in Fig. 8.8, it is evident that increasing NWA temperatures mainly impact the magnitudes of the saturated responses, asat, but hardly the positions of the sensitivity windows. This latter effect is demonstrated in more detail in Fig. 8.10(a) where we have replotted the O2 response data with the maximum response at each temperature scaled to unity. Also shown in this ﬁgure is a series of Langmuir 254 Konrad Maier et al.

(a) (b)

–

R –

–

Figure 8.10 (a) (Data points) Normalized O2 response as observed at nanowire array temperatures ranging in between 25 C and 150 C; (full lines): Langmuir adsorption and recombination isotherms calculated assuming a constant and species-specific oxygen binding energy of ¼ 0.66 eV. (b) E as a function of adsorbent temper- Eads O2 ads ature T. The data points were evaluated from the photoluminescence response data of Fig. 8.8. isotherms which had been calculated based on the assumption that the O2-related recombination centers bind to the InGaN surfaces with a unique fi ¼ and species-speci c adsorption energy of Eads O2 0.66 eV, allowing at the same time the NWA temperature to vary over the full experimental range, i.e., 25 C T 150 C. While this variation in temperature shifts the expected positions of the LAR sensitivity windows by no less than four orders of magnitude, all experimental data approximately fall onto one and the same isotherm corresponding to an adsorbent temperature of about 80C. As this huge discrepancy exceeds any experimental scatter in the determination of PL responses, this effect is discussed in further detail. Mathematically, this discrepancy can be resolved by assuming that the adsorption energy is not a species-dependent constant but rather a temperature-dependent quantity as shown in Fig. 8.10(b). Overall, these data show that the adsorption energies for all species investigated vary linearly with temperature, increasing from values close to zero at cryogenic temperatures to values around 1 eV at temperatures around 150C. Comparing the adsorption energies for the different species at any fixed temperature, it is found that adsorbate-binding energies increase in the order of H2O, O2, EtOH, NO2, and O3. Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 255

At first sight, the data in Fig. 8.10(b) could be interpreted in the way that all investigated adsorbates undergo physisorption at low temperature and remain in this weakly bound state as long as temperatures are low enough to prevent relaxation into more deeply bound chemisorption states. The linear increase in Langmuir adsorption energies at T > 50K further suggests that there are no unique chemisorption states but that there are rather continua of deeper and deeper bound chemisorption states which can be reached at the expense of surmounting higher and higher reaction barriers. Although such a scenario cannot be excluded, we draw attention to a second, more realistic scenario that is able to explain the observed linear increase in adsorption energies with temperature.25 Here, we consider that the InGaN/GaN NWA transducer surfaces are never exposed to a single gas alone and that the different kinds of molecules therefore compete for a limited number of adsorption sites. Competitive adsorption phenomena are well-known and generally recognized in the fields of catalysis and e chromatography.41 44 To the best of our knowledge, however, competitive adsorption has not been seriously considered in the field of gas sensing yet. To introduce the concept of competitive adsorption, we start from the usually considered idealized situation of a single gas interacting with an adsorbent. In such an idealized scenario, Langmuir adsorption leads to a surface coverage q with pgas standing for the gas pressure and T for the e temperature of the adsorbent38 40: pgas q pgas; T; M ¼ : (8.7) Eads pgas þ P00ðT; MÞexp kBT The only parameter in this equation is the adsorption energy Eads which measures the strength of adsorption on the adsorbent under study. As discussed in text books,39 Eq. (8.7) represents the steady-state solution of the differential equation, which describes the kinetics of adsorption: dq pgas; T ¼ r ðTÞ$ 1 q p ; T p r ðTÞ$q p ; T (8.8) dt ads gas gas des gas with rads and rdes standing for the adsorption and desorption rate constants of the analyte in question. In this equation, the term (1 q) considers that any adsorption site on the adsorbent surface can only be occupied once. The idealized case of an adsorbent being exposed to a single gas can be easily generalized. Considering the simplest case of two gas species, A and B, 256 Konrad Maier et al. competing for a single kind of adsorption sites, Eq. (8.8) turns into a system of two coupled equations for the surface coverages qA and qB:

dqA ¼ r ; ðTÞ$ð1 q q Þ$p r ; ðTÞ$q ; (8.9a) dt ads A A B A des A A and

dqB ¼ r ; ðTÞ$ð1 q q Þ$p r ; ðTÞ$q . (8.9b) dt ads B A B B des B B Under steady-state conditions, one then obtains for the surface coverages of adsorbates A and B: pA qA pA; pB;T ¼ ; EA pB EB pA þ P00AðTÞexp 1 þ exp kBT P00BðTÞ kBT (8.10a) and pB qB pA; pB;T ¼ . EB pA EA pB þ P00BðTÞexp 1þ exp kBT P00AðTÞ kBT (8.10b) Both entities, obviously, follow similar isotherms as in the single- adsorbate case. Differences, however, are caused by additional terms that appear in the denominators, which describe the interaction of both adsorbates on the adsorbent surface. Both equations can be reduced to the familiar single-component Langmuir form by replacing the species-dependent adsorption energies in the denominator, EA,B by species-dependent effective adsorption energies, which contain contributions of the competing species as well. Focusing on species A, one obtains EA pB EBEA EA;eff ð pB; TÞ¼ kBT ln exp þ exp ; kBT P00BðTÞ kBT (8.11) and similarly, for EB,eff(pA, T). Obviously, this latter expression does not only depend on the adsorption strength of adsorbate A but it is also sensitive to the difference in adsorption energies of both species. For EA ¼ EB, Eeff,A effectively reduces to Eeff,A ¼ EA as realistic partial pressures pB of component B always remain far smaller than the desorption pressure P00B of species B. Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 257

(a) (b) C

– ( )

Figure 8.11 (a) Variation of the surface coverages of N2 and O2 as the O2 concentration in the N2 background is raised. Data points represent photoluminescence response data obtained on an InGaN/GaN nanowire array kept at a temperature of about 120 C; (b) variation of the effective adsorption energies with temperature for the four investigated air constituents. Data points are the Eads(T) values obtained from Langmuir adsorption and recombination ﬁts.

As an example of competitive adsorption, we show in Fig. 8.11a PL response data of an optimized InGaN/GaN NWA sample with pronounced sensitivity toward O2. There, the observed quenching response of O2 could be followed over three orders of magnitude in O2 concentration and explained as a competition with the more numerous but weaker-binding N2 background molecules, which are supposed to be nonquenching. As shown there, the observed O2 response can be explained by a competition of N2 and O2 adsorbates in case the adsorption energies EO2 and EN2 are ﬁ ¼ ¼ xed at EO2 0.95 eV and EN2 0.75 eV, respectively. According to this result, the N2 adsorbates win this competition if O2 abounds in small concentrations. The InGaN/GaN surface is then almost completely covered with nitrogen. At O2 concentrations around 100 ppm, the stronger-binding O2 molecules start to displace the nitrogen adsorbates, until at approximately 3000 ppm O2, nitrogen and oxygen adsorbates are present in equal concentrations. Finally, at normal ambient air concentrations of O2, the surface is almost completely covered with O2.AsO2 adsorbates tend to adsorb in ionized form, this exchange of N2 for O2 adsorbates obviously can only involve a small fraction of 0.1%e1% of all geometrically available surface states as the Weisz limitation39,40 needs to be respected. ﬁ With the values of EO2 and EN2 being xed by the O2 concentration dependence of Fig. 8.119(a), the temperature dependence of the apparent 258 Konrad Maier et al.

adsorption energy for O2 can be evaluated from Eq. (8.11). With the back- ¼ $ 4 ground nitrogen pressure of pN2 8 10 Pa, the data in Fig. 8.11(b) are obtained. There, the calculated trend for EO2;eff T , is compared to the LAR adsorption energies for O 2 already displayed in Fig. 8.10(b).Consid- ering the calculated EO2;eff T trend, it is seen that the effective binding energy should first increase linearly with temperature and then turn to saturation as the temperature is further raised. Within the much more limited temperature range in which EO2 values could be determined from O2 calibration curves, the predicted linear trend is reasonably well confirmed. Also included in Fig. 8.11(b) are those LAR adsorption data for H2O, EtOH, NO2,andO3 that have already been reported above, alongside with their respective fits to Eq. (8.11). Obviously, the linear increases in LAR adsorption energies for these other gases can also be reasonably well approximated by the competitive adsorption model, provided the high-temperature limits of adsorption energies for these other gases are correctly scaled relative to the initially determined value of EO2.AsEq. (8.11) only depends on the difference in adsorption energies of the respective analyte with its main competitor, these saturation values do not necessarily represent the exact quantum-chemical binding energies of the individual adsorbates. In principle, however, these values would become experimentally accessible in case PL response measurements could be extended into the range of temperatures at which saturation in LAR-binding energies can be observed. Another interesting observation is that the saturation values of Eads determined in Fig. 8.11(b) are linearly correlated to the electron affinity of the test gases (Fig. 8.12). As the electron affinity of molecules measures the energy gained in capturing a free electron on a molecule X, i.e., X þ e/ X ; (8.12) this correlation emphasizes the role of photogenerated electrons in enabling the formation of adsorbate states on InGaN surfaces. Whereas the positive electron affinities of O2,NO2, and O3 mean that free electrons can become exothermally bound to these gases, the negative electron affinities of N2 and H2O mean that electron attachment is endothermic, and that therefore the formation of negatively charged N2 and H2O ions is unlikely. Whereas free water does not attract electrons, electron capture, however, becomes possible when water is disintegrated into H and OH fragments. As already mentioned above, the saturation values of Eads,sat listed in Fig. 8.12 do not necessarily represent the true quantum chemical binding energies of the adsorbates. As any possible shift in the initial value of the Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 259

– – E ( ) 36 Figure 8.12 Saturated values of Eads as a function of the electron afﬁnity of the test gases.

N2-binding energy would shift all other binding energies by the same amount, the correlation in Fig. 8.12 would remain unaltered.

8.6.3 Competition between quenching and enhancing H2O adsorbates The overview presented in Section 8.5 has shown that water vapor influences the PL response of InGaN/GaN NWAs in a more complex manner than most other gases investigated. Whereas O2,NO2, and O3 simply give rise to quenching PL responses, water vapor can form both quenching (Q) and enhancing (E) adsorbates. Most interestingly, our data suggest that initially quenching adsorbates tend to transform into enhancing ones as III-nitride surfaces are continually exposed to water vapor. Given enough time, Q-adsorbates will therefore always tend to transform into E-adsorbates, which represent the equilibrium form of H2O adsorption. In the following, we present evidence that such QeE transformation behavior is another manifestation of a competitive adsorption process that can occur at UV-illuminated III-nitride surfaces. Following the idea that H2O can form two different kinds of competing adsorbates, rate Eqs. (8.9a,b) can be used again, with species A standing for Q- and species B for E-adsorbates. As the competition between both is a time-dependent one, time-dependent solutions to Eqs. (8.9a,b) need to be fitted to experimentally observed PL transients. Such fits then yield values for the adsorption and desorption rate constants of the two kinds of H2O adsorbates as well as activation energies for the kinetic processes of adsorption and desorption. 260 Konrad Maier et al.

The time rates of change of surface coverages qQ and qE ultimately depend on the rate rH2O with which H2O molecules collide with the InGaN surfaces: p r p ; T ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiH2O A (8.13) H2O H2O p ads 2 MH2OkBT

Here, pH2O stands for the H2O partial pressure in the gas phase, MH2O for the water molecular mass, and Aads for the effective area of the adsorption sites. Furthermore, as not every gas-kinetic collision will lead to an adsorption event, we write the two adsorption rates as products of the gas-kinetic fi collision rate, rH2O(pH2O,T), a species-speci c sticking factor (sQ,sE), and a Boltzmann factor containing the species-specific reaction barriers (εads,Q, εads,E): ε ; ¼ ; ads;Q ; rads;Q pH2O T sQrH2O pH2O T exp (8.14a) kBT ε g ; ; ¼ ; ads;E Popt : rads;E pH2O T Popt sErH2O pH2O T exp kBT Popt;max (8.14b) Moreover, as the formation of enhancing adsorbates is accelerated by UV illumination, we introduce in rads,E an additional factor that depends on a power g of the UV input optical power Popt. Similarly, we assume that desorption of both species requires thermal activation as well: εdes;Q rdes;QðTÞ¼ r0;Q exp ; (8.15a) kBT εdes;E rdes;EðTÞ¼ r0;E exp . (8.15b) kBT With the solutions for qQ and qE obtained, the PL response under the influence of water vapor adsorption becomes26 ¼ q RQ þ q RE RPL H2O QA 1 EA 1 (8.16) |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl}Rnr |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl}Rnr

¼:aQ ¼:aE

In this ﬁnal equation, RQ and RE are the nonradiative recombination rates through quenching and enhancing adsorbates, while Rnr is the nonradiative recombination rate under clean-air conditions; A is a common scale factor. Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 261

In Fig. 8.13(a), we show measurements of the water vaporeinduced PL response as measured at a relative humidity of 30%, but at different NWA temperatures and at different LED light exposure levels. In Fig. 8.13(b), these results are compared with those PL responses that were calculated using the above set of equations. For simplicity, we assumed ideal boxlike H2O exposure pulses as shown in the bottom panels in Fig. 8.13. As can be seen, the simulated PL data reasonably well reproduce the observed features of the actual PL response. Water vapor exposure pulses, in particular, can be seen to produce initially quenching responses which more or less rapidly turn into enhancing ones as the exposures are being maintained. Another encouraging aspect is that the PL overshoots that occur after the termination of each H2O exposure pulse can also be reproduced. More insight into the underlying microscopic processes can be gained by looking at those values of model parameters that are revealed by the above fits. Turning to the adsorption parameters first, we find that the sticking 2 probabilities for forming Q-adsorbates are quite low (sQ¼3.5$10 ), and 6 even lower for E-adsorbates (sE¼2.4$10 ). More interestingly, the formation of Q-adsorbates does not seem to require thermal energy (εads,Q z 0 eV per adsorbate), while E-ones require sizable amounts of activation energy (εads,E z 0.4 eV per adsorbate). Furthermore, the formation of E-adsorbates is found to increase with the square root of the UV LED power, i.e., g ¼ 1/2. Another interesting result is that both kinds of adsorbates seem to desorb without any significant input in thermal energy (εdes,Qy εdes,E z 0 eV). Both kinds of adsorbates, however, desorb at very different temperature-independent rates (rdes,Q z 10 Hz vs. rdes,E z 5$10 2 Hz). Overall, our kinetic considerations show that there is hardly any energy involved in the adsorption and desorption of quenching adsorbates, whereas sizable amounts of thermal energy and, even more UV light energy, are required for the formation of enhancing adsorbates. It is therefore reasonable to associate Q-adsorbates with physisorbed and E-adsorbates with chemisorbed water molecules, the dividing line between physi- and chemisorption being the typical strength of a hydrogen bridge 36 bond (EH z 0.4 0.5eV).

8.7 Molecular mechanism of adsorption Our experiments have revealed that molecular adsorption at InGaN surfaces can both quench and enhance the native PL of InGaN/GaN NWAs. The fact that adsorption can modify the native PL in both directions 262 Konrad Maier et al.

(a)M (b) S

P opt = 0.7 mW P opt = 0.7 mW

Popt = 200 mW Popt = 200 mW ) (

T ( ) T ( ) Figure 8.13 (a) Experimental photoluminescence (PL) response data as observed under widely different experimental conditions. For each level of excitation-light intensity, the variation of the PL-light intensity in response to the humidity profiles (gray-filled boxes) is shown for three different temperatures; (b) values of PL response as obtained from the competitive adsorption model. suggests that the native PL is limited by intrinsic surface sites and that their trapping and recombination cross sections are being altered as adsorbates bind to them. Our gas sensing tests have revealed equilibrium adsorption energies Eads ¼ εdes εads of several groups of adsorbates on these sites and some additional information about the kinetic parameters εads and εdes has been obtained in the case of water adsorption. The microscopic nature of the intrinsic surface sites and the ways in which adsorbates bind to these sites, however, have not yet been positively identified. Identification of these molecular entities clearly remains a subject of future research. A possible approach to attain such information is to apply diffuse reflection Fourier transform spectroscopy (DRIFTS)45,46 to InGaN/GaN NWAs as these are exposed to different analytes. To motivate such research, we close this chapter with some ideas concerning the possible nature of the intrinsic surface sites and about the ways adsorbates might bond to these. Starting point of our considerations is the nature of chemical bonds at nonpolar III-nitride surfaces. The bulk equilibrium lattice sites of Ga(In) 38e40 and N, which in Kroger€ eVink notation are denoted by GaGa (InIn) and NN, all feature tetrahedral coordination. Within the bulk, the tetrahedral coordination of the constituent atoms of InGaN/GaN nanowires is enabled by an electron transfer from N toward Ga or In atoms. In contrast Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 263 to the Kroger€ eVink notation in which all ground state equilibrium configurations have formal charge zero, we denote these bulk equilibrium sites by þ Ga4 (In4 ) and N4 . In this notation, which has become known under different names such as octet or 8-N rules,47,48 the superscript () denotes the ionic charge of the constituent atoms and the subscript the number of covalent bonds to neighbor atoms required to arrive either at completely filled or empty valence shells. The advantage of the 8-N notation is that it most visibly keeps track of those valence electron transfers that take place during electronically driven coordination changes. In the realm of solid-state physics, 8-N considerations have proved to be useful to understand electronically driven coordination changes in the bulk of amorphous e semiconductors.48 50 Here, we employ such considerations to the analysis of electronically driven coordination changes at molecules adsorbed on photoexcited semiconductor surfaces. Moving from the bulk of InGaN/GaN nanowires to their surfaces, the tetrahedral coordination of the bulk atoms cannot be continued. Due to a lack of bond partners, the InGaN/GaN constituents cannot form four covalent bonds at the surface, which results either in a high density of dangling bonds at the surface or in the formation of threefold coordinated N-Ga(In) sites with reconstructed surface bonds. Following the above notation, these 0 0 0 threefold coordinated surface sites can be labeled as N3,Ga3,andIn3.A pair of such reconstructed sites at a nonpolar 1100 lateral nanowire surface is sketched in Fig. 8.14(a). In principle, such a pair can transform from its reconstructed ground state into an activated excited state once a photogenerated electronehole pair becomes trapped at this pair. The transformation

(a) / (b) B A

0 0 + (Ga –) (N ) N Ga (Ga3 ) (N4 ) N Ga 4 3 P + –

– – + (Ga4 ) Ga N (Ga4 ) Ga N (N4 ) R G P Figure 8.14 (a) Tetrahedral bulk bonding and surface reconstruction at lateral 1100 surfaces. The coloring indicates formation of Lewis acidebase pairs; (b) photoactivated state of III-nitride surface following photogeneration in the bulk. The circles with plus and minus signs denote a photogenerated electronehole pair. 264 Konrad Maier et al. shown in Fig. 8.14(b) requires that an electronehole pair, initially delocalized across the entire InGaN well volume, becomes sharply localized at a pair of Ga(In) and N atoms. Such localization is necessary to make the trapped electronic charge chemically active in the sense of enabling local redox reactions as those shown in Fig. 8.14. Such localization is a relatively improbable process. In amorphous semiconductors, such localization phenomena occur in 3 the small fraction floc < 10 of localized band-tail states which is a small fraction of the total number of electronic band states. The other requirement for enabling local coordination changes is the lack of geometrical constraints that would otherwise prevent electronically destabilized chemical bonds to disintegrate and to reform in a different manner. In amorphous semiconductors, local coordination changes are enabled by density-deficient regions like e voidlike defects and/or by the diffusional motion of bonded hydrogen.48 50 All such changes are clearly impossible within the bulk of a fully coordinated crystalline material as in the interior of an InGaN/GaN nanowire. A crystalline sensor surface, however, is a far less constrained environment. We therefore conjecture that a small fraction of the total number of Ga(In)-N surface sites may actually support local transformations as those shown in Fig. 8.14(b). In gas sensors and in heterogeneous catalysis, the number density of charged surface sites is limited by the Weisz limitation.39,40 Physically, this limitation arises because surface charge densities in the order of 1012cm 2 will generate electrical fields comparable to the breakdown field of the underlying semiconductor. This well-established limit suggests that again only a small 3 fraction fWZL < 10 of all Ga(In)-N surface sites will actually be able to support bond reconfigurations as those shown in Fig. 8.14. Accepting such a possibility of local reconfiguration, the following picture emerges: turning to the reconstructed ground state in Fig. 8.14(a) first, we note that N being a potential electron donor and Ga an electron acceptor, a surface N-Ga pair can be considered as a Lewis acid/base (LAB) pair. On photoexcitation (hn) such pairs can trap electronehole pairs þ as shown in Fig. 8.14b forming N4 and Ga4 (or In4 , respectively): 0 þ 04 þ þ ; N3 Ga3 N4 Ga4 (8.17) When such a process occurs in vacuum or in an inert atmosphere, the pair of charged dangling bonds will discharge after a time sr as the electron e þ trapped on the Ga4 site tunnels back to its neighboring N4 site. When the energy difference between initial and final states can be dissipated in the form of phonons, the ensuing recombination process will be radiation-less and result in PL quenching. Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 265

(a) (b) B A

0 0 + (O2 ) Ga O Ga (Ga ) P – 3 (O3 ) Ga O Ga (Ga4 ) + – N Ga N (N +) + 4 N Ga N (N4 ) R G P

Figure 8.15 (a) Lewis acidebase pair formed by Ga and embedded oxygen at lateral 1100 surfaces; (b) photoactivated state of an oxidized III-nitride surface after photoexcitation.

As InGaN/GaN surfaces exposed to ambient air are likely to form thin e layers of natural oxide,15 21 similar reconfigurations are also conceivable on oxidized surfaces (Fig. 8.15). Like on nonoxidized surfaces, pairs of neighboring oxygen and gallium atoms can form LAB pairs which can switch between reconstructed ground states and electronically excited states with two dangling bond radicals pointing out of the surface: 0 þ 04 þ þ O2 Ga3 O3 Ga4 (8.18) When InGaN/GaN NWAs are operated in backgrounds containing oxygen or other kinds of reactive gases, the charged dangling bonds pointing out of electronically excited sensor surfaces can form cross-linking bonds and thus allow reactive gases to form adsorbates. Such chemisorption bonds alter the trapping and recombination cross sections of the native LAB pairs and thus promote changes in the luminescence output which become experimentally observable in the form of a gas response. These latter possibilities are sketched in Fig. 8.16, indicating that adsorbate bonding can both quench (O2,NO2, and O3) and enhance (H2O) the native PL response. The above considerations about adsorbate bonding describe a possible scenario that can account in a qualitative manner for the gas response data described in earlier sections of this chapter. As adsorbate bonding is a complex and exciting field of research with many unresolved issues and challenges ahead, we expect that this picture will become modified as additional spectroscopic evidence becomes available. A key enabling factor for attaining such information is that the PL probing of gas response can potentially be combined with other forms of operando spectroscopy, in particular DRIFTS.45,46 266 Konrad Maier et al.

(a) (b) (c) (d) O W w a e H O O G P H H – O H + – + + –

N Ga N Ga N Ga N Ga

Ga N Ga N Ga N Ga N

Figure 8.16 Lewis acidebase (LAB) pair alternating between ground (a) and electronically excited state (b) giving rise to thermal quenching in vacuum or in inert gas atmospheres; (c) LAB modiﬁed by dissociative oxygen adsorption featuring an enhanced nonradiative recombination rate; and (d) LAB with surface bonds passivated by water fragments featuring a reduced nonradiative recombination rate.

8.8 Conclusions and outlook InGaN/GaN nanowire structures proved to be very efficient nanooptical probes for investigating adsorption processes on semiconductor surfaces. Because PL emission does not depend in a similarly complex way on the transport of photoexcited electronehole pairs as in resistively readout gas sensors, luminescence probing provides a more direct approach to the observation of adsorption processes on semiconductor surfaces. Looking toward future perspectives, improvements in material’s quality and the application of nanowire heterostructures with strong carrier confinement as well as refinements in the optical readout periphery will be able to extend the accessible temperature range of the optochemical transducers to temperatures up and beyond the 200 C range where resistively readout MOX gas sensors are routinely operated and where thermally activated surface oxidation and surface combustion processes take place. As combinations of PL response with DRIFTS measurements appear to be experimentally feasible, PL adds another useful tool to the toolbox of in operando spectroscopies.45,46 We are therefore confident that InGaN/GaN NWA will provide deeper insights into the microscopic processes underlying adsorption at oxide and nonoxide surfaces in the near future. Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 267

Concerning future applications in the field of gas sensors, it should be kept in mind that InGaN/GaN NWAs can also be grown on transparent sapphire or GaN substrates. In this way, the electrical readout periphery can be completely removed from the spot of sensing where the sensitive materials are in contact with the medium to be sensed. By employing light fibers for carrying UV excitation light to the NWAs and luminescence light back to the light detector, InGaN/GaN nanooptical probes can also be operated in environments heavily affected by electromagnetic interference.14 Another interesting aspect is that by means of selective area growth, III-nitride nanowires can be grown in the form of regular arrays with wire-to-wire distances in the order of visible light wavelengths. With the help of such sparse arrays not only a better media access is enabled but also a more efficient light coupling can be achieved by exploiting photonic crystal effects.51 References 1. Moseley PT, Norris JOW, Williams DE. Techniques and mechanisms in gas sensing. Adam Hilger; 1991. 2. Ihokura K, Watson J. The stannic oxide gas sensor principles and applications. Boca Raton: CRC Press; 1994. 3. Korotcenkov G. Handbook of gas sensor materials; properties, advantages and shortcomings for applications volume 1: conventional approaches. 2013. 4. Comini E, Faglia G, Sberveglieri G. Solid state gas sensing. 1st ed. Boston, MA: Springer Science & Business Media; 2009. https://doi.org/10.1007/978-0-387-09665-0. 5. Yamazoe N, Shimanoe K. Receptor function and response of semiconductor gas sensor. J Sensors 2009;2009:1e21. https://doi.org/10.1155/2009/875704. 6. Hua Z, Li Y, Zeng Y, Wu Y. A theoretical investigation of the power-law response of metal oxide semiconductor gas sensors I: Schottky barrier control. Sensor Actuator B Chem 2018;255:1911e9. https://doi.org/10.1016/J.SNB.2017.08.206. 7. Hua Z, Qiu Z, Li Y, Zeng Y, Wu Y, Tian X, Wang M, Li E. A theoretical investigation of the power-law response of metal oxide semiconductor gas sensors II: size and shape effects. Sensor Actuator B Chem 2018;255:3541e9. https://doi.org/10.1016/J.SNB. 2017.09.189. 8. Ulrich M, Bunde A, Kohl C-D. Percolation and gas sensitivity in nanocrystalline metal oxide films. Appl Phys Lett 2004;85:242e4. https://doi.org/10.1063/1.1769071. 9. Sauerwald T, Russ S. Percolation effects in metal oxide gas sensors and related systems. In: Kohl C-D, Wagner T, editors. Gas sens. Fundam. Berlin Heidelberg: Springer; 2013. https://doi.org/10.1007/5346. 10. Faglia G, Baratto C, Sberveglieri G, Zha M, Zappettini A. Adsorption effects of NO2 at ppm level on visible photoluminescence response of SnO2 nanobelts. Appl Phys Lett 2005;86:011923. https://doi.org/10.1063/1.1849832. 11. Comini E, Baratto C, Faglia G, Ferroni M, Sberveglieri G. Single crystal ZnO nanowires as optical and conductometric chemical sensor. J Phys D Appl Phys 2007; 40:7255e9. https://doi.org/10.1088/0022-3727/40/23/S08. 12. Pallotti DK, Passoni L, Gesuele F, Maddalena P, Di Fonzo F, Lettieri S. Giant O2 -induced photoluminescence modulation in hierarchical titanium dioxide nanostructures. ACS Sens 2017;2:61e8. https://doi.org/10.1021/acssensors.6b00432. 268 Konrad Maier et al.

13. Palmer DW. Properties of III-nitride semiconductors. 2018. http://www.semiconductors. co.uk/nitrides.htm. 14. Paul S, Helwig A, Muller€ G, Furtmayr F, Teubert J, Eickhoff M, Sumit P. Opto-chemical sensor system for the detection of H2 and hydrocarbons based on InGaN/GaN nanowires. Sensor Actuator B Chem 2012;173:120e6. https://doi.org/ 10.1016/j.snb.2012.06.022. 15. Schalwig J, Muller€ G, Karrer U, Eickhoff M, Ambacher O, Stutzmann M, Gorgens€ L, Dollinger G. Hydrogen response mechanism of PteGaN Schottky diodes. Appl Phys Lett 2002;80:1222e4. https://doi.org/10.1063/1.1450044. 16. Weidemann O, Hermann M, Steinhoff G, Wingbrant H, Lloyd Spetz A, Stutzmann M, Eickhoff M. Inﬂuence of surface oxides on hydrogen-sensitive Pd:GaN Schottky diodes. Appl Phys Lett 2003;83:773. https://doi.org/10.1063/1.1593794. 17. Shalish I, Shapira Y, Burstein L, Salzman J. Surface states and surface oxide in GaN layers. J Appl Phys 2001;89:390e5. https://doi.org/10.1063/1.1330553. 18. Garcia MA, Wolter SD, Kim T-H, Choi S, Baier J, Brown A, Losurdo M, Bruno G. Surface oxide relationships to band bending in GaN. Appl Phys Lett 2006;88:013506. https://doi.org/10.1063/1.2158701. 19. Winnerl A, Garrido JA, Stutzmann M. GaN surface states investigated by electrochemical studies. Appl Phys Lett 2017;110:101602. https://doi.org/10.1063/1.4977947. 20. Wang L, Bu Y, Li L, Ao J-P. Effect of thermal oxidation treatment on pH sensitivity of AlGaN/GaN heterostructure ion-sensitive ﬁeld-effect transistors. Appl Surf Sci 2017; 411:144e8. https://doi.org/10.1016/j.apsusc.2017.03.167. 21. Kehagias T, Dimitrakopulos GP, Becker P, Kioseoglou J, Furtmayr F, Koukoula T, H€ausler I, Chernikov A, Chatterjee S, Karakostas T, Solowan H-M, Schwarz UT, Eickhoff M, Komninou P. Nanostructure and strain in InGaN/GaN superlattices grown in GaN nanowires. Nanotechnol 2013;24:435702. https://doi.org/10.1088/ 0957-4484/24/43/435702. 22. Teubert J, Becker P, Furtmayr F, Eickhoff M. GaN nanodiscs embedded in nanowires as optochemical transducers. Nanotechnol 2011;22:275505. https://doi.org/10.1088/ 0957-4484/22/27/275505. 23. Teubert J, Paul S, Helwig A, Muller€ G, Eickhoff M. Group III-nitride chemical nanosensors with optical readout. In: Kohl C-D, Wagner T, editors. Gas sens. Fundam. Springer Berlin Heidelberg; 2014. p. 311e38. https://doi.org/10.1007/ 5346_2014_58. 24. Maier K, Helwig A, Muller€ G, Becker P, Hille P, Schormann€ J, Teubert J, Eickhoff M. Detection of oxidising gases using an optochemical sensor system based on GaN/ InGaN nanowires. Sensor Actuator B Chem 2014;197:87e94. https://doi.org/ 10.1016/j.snb.2014.02.002. 25. Maier K, Helwig A, Muller€ G, Hille P, Teubert J, Eickhoff M. Competitive adsorption of air constituents as observed on InGaN/GaN nano-optical probes. Sensor Actuator B Chem 2017;250:91e9. https://doi.org/10.1016/j.snb.2017.04.098. 26. Maier K, Helwig A, Muller€ G, Hille P, Teubert J, Eickhoff M. Photoluminescence probing of complex H2O adsorption on InGaN/GaN nanowires. Nano Lett 2017;17: 615e21. https://doi.org/10.1021/acs.nanolett.6b03299. 27. Maier K, Helwig A, Muller€ G, Eickhoff M. Photoluminescence detection of surface oxidation processes on InGaN/GaN nanowire arrays. ACS Sens 2018;3(11): 2254e60. https://doi.org/10.1021/acssensors.8b00417. 28. She X, Huang AQ, Lucia O, Ozpineci B. Review of silicon carbide power devices and their applications. IEEE Trans Ind Electron 2017;64:8193e205. https://doi.org/ 10.1109/TIE.2017.2652401. Luminescence probing of surface adsorption processes using InGaN/GaN nanowire 269

29. Rath P, Ummethala S, Nebel C, Pernice WHP. Diamond as a material for monolithically integrated optical and optomechanical devices. Phys Status Solidi 2015;212: 2385e99. https://doi.org/10.1002/pssa.201532494. 30. Flack TJ, Pushpakaran BN, Bayne SB. GaN technology for power electronic applications: a review. J Electron Mater 2016;45:2673e82. https://doi.org/10.1007/s11664- 016-4435-3. 31. Chao E. Analysis of LED technologies for solid state lighting markets e technical report No. UCB/EECS-2012-138. 2012. https://www2.eecs.berkeley.edu/Pubs/TechRpts/ 2012/EECS-2012-138.html. 32. Ambacher O, Majewski J, Miskys C, Link A, Hermann M, Eickhoff M, Stutzmann M, Bernardini F, Fiorentini V, Tilak V, Schaff B, Eastman LF. Pyroelectric properties of Al(In)GaN/GaN hetero- and quantum well structures. J Phys Condens Matter 2002; 14:3399e434. https://doi.org/10.1088/0953-8984/14/13/302. 33. Fletcher ASA, Nirmal D. A survey of gallium nitride HEMT for RF and high power applications. Superlattice Microst 2017;109:519e37. https://doi.org/10.1016/ J.SPMI.2017.05.042. 34. Pearton SJ, Kang BS, Kim S, Ren F, Gila BP, Abernathy CR, Lin J, Chu SNG. GaN- based diodes and transistors for chemical, gas, biological and pressure sensing. J Phys Condens Matter 2004;16:R961e94. https://doi.org/10.1088/0953-8984/16/29/R02. 35. Pearton SJ, Ren F. Gallium nitride-based gas, chemical and biomedical sensors. IEEE Instrum Meas Mag 2012;15:16e21. https://doi.org/10.1109/MIM.2012.6145256. 36. NIST. Chemistry webBook. 2016. http://webbook.nist.gov/chemistry/. 37. Wu T-Y. Electron affinity of boron, carbon, nitrogen, and oxygen atoms. Phys Rev E 1955;100:1195e6. https://doi.org/10.1103/PhysRev.100.1195. 38. Kittel C, Kroemer H. Thermal physics. 2nd ed. New York: W. H. Freeman and Company; 1998. 39. Henzler M, Gopel€ W. Oberfl€achenphysik des Festk€orpers. 2nd ed. Wiesbaden: Teubner Verlag; 1994. 40. Morrison SR. The chemical physics of surfaces. Boston, MA: Springer US; 1977. https:// doi.org/10.1007/978-1-4615-8007-2. 41. Gun’ko VM. Competitive adsorption. Theor Exp Chem 2007;43:139e83. https:// doi.org/10.1007/s11237-007-0020-4. 42. Martin RJ, Al-Bahrani KS. Adsorption studies using gas-liquid chromatographydII. Competitive adsorption. Water Res 1977;11:991e9. https://doi.org/10.1016/0043- 1354(77)90157-9. 43. Jacobson JM, Frenz JH, Horvath CG. Measurement of competitive adsorption isotherms by frontal chromatography. Ind Eng Chem Res 1987;26:43e50. https:// doi.org/10.1021/ie00061a009. 44. Poplewska I, Pia˛tkowski W, Antos D. Effect of temperature on competitive adsorption of the solute and the organic solvent in reversed-phase liquid chromatography. J Chromatogr A 2006;1103:284e95. https://doi.org/10.1016/j.chroma.2005.11.038. 45. Weckhuysen BM. Operando spectroscopy: fundamental and technical aspects of spectroscopy of catalysts under working conditions. Phys Chem Chem Phys 2003;5:1. https://doi.org/10.1039/b309654h. 46. Degler D, Barz N, Dettinger U, Peisert H, Chassé T, Weimar U, Barsan N. Extending the toolbox for gas sensor research: operando UV/vis diffuse reflectance spectroscopy on SnO2-based gas sensors. Sensor Actuator B Chem 2016;224:256e9. https://doi.org/ 10.1016/J.SNB.2015.10.040. 47. Langmuir I. The arrangement of electrons in atoms and molecules. J Am Chem Soc 1919; 41:868e934. https://doi.org/10.1021/ja02227a002. 270 Konrad Maier et al.

48. Mott NF. Electrons in disordered structures. Adv Phys 1967;16:49e144. https:// doi.org/10.1080/00018736700101265. 49. Muller€ G, Kalbitzer S, Mannsperger H. A chemical-bond approach to doping, compensation and photo-induced degradation in amorphous silicon. Appl Phys A Solid Surf 1986;39:243e50. https://doi.org/10.1007/BF00617268. 50. Robertson J. Mott lecture: how bonding concepts can help understand amorphous semiconductor behavior. Phys Status Solidi 2016;213:1641e52. https://doi.org/ 10.1002/pssa.201532875. 51. Winnerl J, Hudeczek R, Stutzmann M. Optical design of GaN nanowire arrays for photocatalytic applications. J Appl Phys 2018;123:203104. https://doi.org/10.1063/ 1.5028476. CHAPTER NINE

Rareearthedopedoxidematerials for photoluminescence-based gas sensors

V. Kiisk, Raivo Jaaniso University of Tartu, Tartu, Estonia

Contents

9.1 Introduction 272 9.1.1 The concept of PL-based gas sensing 272 9.1.2 Advantages of rare eartheactivated inorganic sensor materials 273 9.1.3 Overview of the progress 275 3þ 9.2 Sm :TiO2 277 9.2.1 Introduction 277 9.2.2 Preparation and characterization of samples 278 9.2.3 PL-based oxygen sensing 280 9.2.4 Sensing mechanism and its mathematical model 282 3þ 9.2.5 Multivariable sensing with TiO2:Sm 286 3þ 9.3 Eu :ZrO2 288 9.3.1 Introduction 288 9.3.2 Preparation and characterization of samples 289 9.3.3 Oxygen sensing 290 9.3.4 Sensing mechanism 291 3þ 9.4 Tb :CePO4 294 9.4.1 Introduction 294 9.4.2 Preparation and characterization of samples 295 9.4.3 Gas sensing and its mechanism 295 3þ 9.5 Pr :(K0.5Na0.5)NbO3 298 9.5.1 Introduction 298 9.5.2 Synthesis 298 9.5.3 Oxygen sensing 298 9.6 Conclusion 299 References 300

9.1 Introduction 9.1.1 The concept of PL-based gas sensing Electrical (such as electrochemical or conductometric) gas sensors are widely used, but in many cases, optical sensing has distinct advantages. Not only it is more tolerant to electrical noise and independent of material’s electrical properties (including their long-term drifts due to, e.g., growth of contacting necks between the particles in granular materials) but potentially allows remote access to the sensing volume and operation in imaging mode.1 Fluorescence, or more generally photoluminescence (PL), is one of the most sensitive optical signals (compared with absorption or Raman scattering), although usually not from the gas molecules themselves but from some dedicated PL centers in the solid, which either directly or indirectly interact with the gas molecules. In addition, as opposed to absorption, PL is a two-step process (excitation and emission), so that external inﬂuences have more chances to intercept the emission of a photon. Conventional chemiresistive gas sensing involves measurement of the electrical conductance (or resistance) of a thin semiconductor layer, typically ametaloxide(MOX)ﬁlm (Fig. 9.1). Usually, the material is simultaneously heated and/or optically stimulated to decrease the response time (e.g., stimulate recovery from gas adsorption). The setup for luminescence-based sensing looks somewhat similar, but instead of electrical conductivity, one measures the intensity of the secondary emission from the material (i.e., PL). In this case, photostimulation (also called photoexcitation) is mandatory. Note that, while the electrical conductivity is an integrated quantity characterizing the bulk of the material, the optical signal could

Photostimulation (optional) Photostimulation

Luminescence

Sensor film

Heater Heater

A Figure 9.1 The principle of luminescence-based gas sensing (right) compared with that of conductometric sensing (left). Rare earthedoped oxide materials for photoluminescence-based gas sensors 273 be detected (in principle) separately from each point of the illuminated material. As a well-known example, optical oxygen sensing is frequently done by utilizing oxygen-sensitive PL probes embedded into an oxygen permeable polymer or porous solegel matrix. These probes are mostly organic molecules with energy levels resonant with the triplet oxygen levels and a long excited state lifetime.1 Accordingly, the sensing effect is based on PL quenching via the so-called collisional energy transfer process between the probe and the oxygen molecules. Hence, the fluorescence intensity and its decay time monotonically decrease with increasing oxygen concentration (partial pressure) in a predictable manner as quantified by the SterneVolmer relationship. Suitable selection of the probe allows sensitive operation over a specific (partial) pressure range of the target gas. Equivalently, one can also consider the fluorophores as pressure sensors.2 In principle, the PL signal can be collected from tiny volumes (compared with the volume required for infrared optical absorption by gas molecules3). The system can be rather tightly integrated using miniature solid-state excitation sources and detectors. In this regard, luminescent gas sensing is comparable and compatible with the chemiresistive gas sensing. Semiconductors can exhibit simultaneous gas-sensitive electrical conductivity and PL signals. Moreover, solid matrices can accommodate several different PL centers (or several emissive transitions of single PL center), which can be distinguished spectrally. These electrical and optical signals can complement each other providing further possibilities to improve the specificity, accuracy, or dynamic range of gas sensors. PL can also be employed as a sensitive analytical tool for probing sophisticated processes initiated by surface reactions in gas sensors or photocatalysts.4,5 9.1.2 Advantages of rare eartheactivated inorganic sensor materials There are numerous organic luminescent probes available, mainly based on polycyclic aromatic hydrocarbons and metaleligand complexes.1 However, organic fluorophores are susceptible to photobleaching and cannot with- stand temperatures beyond a few hundred centigrades (at least not in a chemically reactive environment). Such conditions are frequently encountered in industrial applications (for example, in the thermal treatment of polymer packaging for food or healthcare products). Thermal quenching of the fluorescence is sometimes evident already at room temperature. There is accordingly an interest to develop inorganic gas-sensitive luminophores 274 V. Kiisk and Raivo Jaaniso which show stable operation in more aggressive environments and under intense optical excitation. In nanocrystalline form, many inorganic phosphors exhibit certain ambient sensitivity. For example, recent studies revealed that the intensity of the intrinsic luminescence of many common MOX nanopowders (ZnO, TiO2, SnO2,WO3) notably reacts to the change of ambient environment.6,7 Other works have attempted to harness the defect-related or excitonic PL (mostly of ZnO, SnO2, TiO2, and MgO) using a more refined e morphology of the nanomaterial, such as nanostructured films8 10 or nano- e wires/-rods/-belts,11 17 occasionally achieving ppm-level detection of haz- e ardous gases9,11 16 or sensing of oxygen over a wide pressure range.8,17,18 However, the presence and emission properties of lattice defects (such as F-centers) and excitons can be drastically dependent on the quality of the material. On the other hand, such dielectric or semiconductor matrices with relatively wide energy gap can accommodate various impurities acting as PL centers with well-defined luminescence properties. Of the several distinct classes of impurity ions, trivalent rare earth (RE) ions constitute highly pho- tostable impurity centers possessing predictable narrow excitation and emission bands and long fluorescence lifetimes (w1 ms).19,20 Especially in a regular crystalline surrounding, the RE ions exhibit a series of sharp spectral lines typical for the 4fe4f transitions. These features simplify the detection of the sensor signal, monitoring either PL decay kinetics or PL intensities at several different emission wavelengths resulting in a ratiometric response. Because of peculiarities of their energy level schemes, the main emission þ þ transition of several RE ions (such as Eu3 and Tb3 ) is also quite resistant to cross relaxation and thermal quenching. The main quenching mechanism, multiphonon relaxation, becomes apparent only at relatively high temperatures (assuming low phonon frequencies of the host medium21). This is also the basis for the use of RE-activated refractory materials for optical sensing of high temperatures, such as encountered in the thermal barrier coatings of gas turbines.20 In many cases of the studied RE-doped materials (as reviewed in the following), it is believed that the gas sensing stems from a redox reaction, i.e., charge transfer between some lattice ions or defects (including the activator itself) and the adsorbate molecules. In particular, the sensing mechanism may involve trapping or release of free charge carriers in the energy bands of the semiconductor matrix leading effectively to a long range interaction. This is necessary as the PL center is usually located inside the Rare earthedoped oxide materials for photoluminescence-based gas sensors 275 nanocrystal. The chemical change induced by gas adsorption can intercept either the excitation or emission path (or both) of a PL center. Unlike the oxygen sensors based on direct quenching of PL by gas molecules, this mechanism can cause either increase or decrease of the PL intensity as the material is exposed to the gas. The excitation path can be easily affected only if the luminescence is sensitized (through energy transfer from the host or other impurities). There are also inorganic materials where luminescence itself originates from the surface (such as the recombination of electronehole pairs in some semiconductors), and the emission center can more directly interact with the adsorbed or even the gaseous oxygen. For example, Nagai and Noguchi reported already in 1978 that the PL of cleaved n-InP (110) surface 22 reversibly reacted to O2,H2,N2, and H2O. More recently, oxygen was found to have a remarkably strong quenching effect on the PL attributed to radiative decay of excitons localized at the corners or edges of MgO nanocubes.18,23 For InGaN/GaN nanowires, this type of gas response is described in Chapter 8. 9.1.3 Overview of the progress

One of the earliest works studied porous Eu2O3-gAl2O3 composites (powder compacts), where the 325 nm laser irradiation in vacuum decreased the þ Eu3 PL intensity while the same laser irradiation in oxygen gas recovered the PL.24 Such kind of response clearly opposes that of organic fluorophores. þ þ The effect was attributed to Eu3 /Eu2 valence change coupled to creation or annihilation of oxygen vacancies at gAl2O3 and Eu2O3 particle surfaces. The transitions were clearly photoinduced, so that the material was proposed as erasable photomemory. At room temperature, under a laser power density of just 32 mW/cm2, the response time was less than 10 min. þ One of the most interesting gas-sensitive RE-activated material, Sm3 - 25 doped nanocrystalline TiO2 (anatase), was first reported in 2005 and was more rigorously studied in recent years in the form of solegel-prepared e þ nanopowder.7,26 28 PL of the Sm3 ions was found to be reasonably responsive to oxygen gas even at room temperature (with a response time of a few minutes). Similarly to Eu2O3-gAl2O3, the PL significantly improved in an oxygen-rich environment, but the mechanism is quite different and based on an indirect influence of adsorbed oxygen on the fluo- þ rescence quantum yield of Sm3 . Moreover, the ultraviolet excitation is initially absorbed by the matrix which thereafter transfers the energy to the RE emission center,29,30 further complicating the interpretation of the 276 V. Kiisk and Raivo Jaaniso sensing mechanism.28 It is remarkable that the PL is oxygen-sensitive over a pressure range spanning at least four orders of magnitude and reaching the trace concentrations. 3þ Recently, somewhat similarly prepared TiO2:Eu nanopowder was þ reported also exhibiting a decreased Eu3 PL (under 325 nm excitation) þ with reduced ambient oxygen pressure.31 However, in this case, the Eu3 emission lines were relatively broad and there remains a possibility that the ions are prevalently located at the surface. Most recent studies (unpub- 3þ lished data) have shown that the Sm :TiO2 PL is quite sensitive not only to 3þ oxygen but also to NH3 and H2O. More interestingly, PL of Nd :TiO2 appears to exhibit a similarly strong, but reversed oxygen-sensing behavior, where the mechanism is quite different and connected to the excitation efficiency, which is again indirectly affected by gas adsorption. The similar quenching effect of oxidizing gases (O2 or NO2) has been already reported 4,8 9,10,16 for the intrinsic (band-to-band excited) PL of TiO2, ZnO, and 11,16 SnO2 nanostructured films or nanobelts. Microspectroscopic studies of rather small (mean diameter w10 nm) 3þ crystallites of TiO2:Eu demonstrated a case where a significant fraction þ of the RE ions were located at surface.32 The surface and interior Eu3 cen- 5 7 ters could be spectroscopically distinguished by the ratio of the D0/ F1 5 7 and D0/ F2 transitions, because the latter is hypersensitive to local symmetry. While exciting the nanoparticles with 405 nm laser in argon atmosphere, activation of certain light-emitting defect sites (presumably formed þ at surface) was observed. The ratio of the two Eu3 emission bands was reversibly changed as well. It was proposed that energy transfer from free þ and trapped excitonic states to Eu3 ions takes place, but the trapped exci- þ tons (at the defect sites) can only excite surface-located Eu3 ions. In 2010, a patent was issued claiming multiple methods of optical oxygen þ sensing based on the measurements of the PL intensity of Eu3 ions doped in 33 ZrO2 nanocrystallites not bigger than 60 nm. Working temperature was proposed in the range of 0e350C. Recently also several peer-reviewed articles were published reporting similar sensing effect from solegel-prepared 3þ 34,35 ZrO2:Eu particles operated at 300 C. The studies showed that, at 3þ 3þ least for low Eu concentrations, the response mechanism of ZrO2:Eu 3þ is quite similar to that of TiO2:Sm . Interestingly, it was possible to control the magnitude and even the sign of the relative PL response by codoping þ þ with niobium, where presumably the combined effect of Nb5 and Eu3 ions controls the number of vacancies or other charge-compensating defects in the material.34,35 Rare earthedoped oxide materials for photoluminescence-based gas sensors 277

Similar kind of oxygen sensing mechanism was more explicitly demon- 3þ 36 strated on the basis of CePO4:Tb nanocrystals. This material utilizes the fact that several RE ions in solids can exist as redox couples, in this case þ þ Ce3 /Ce4 . The change of valence was induced by exposing the material to the oxidizing or reducing atmosphere at 200C. The activator ion þ (Tb3 ) did not change its valence, but both its excitation and emission properties were affected by the valence change of cerium. A complementary case of PL-based oxygen sensing was demonstrated 3þ 37 with Pr -doped (K0.5Na0.5)NbO3. In addition to the response observed in the absolute PL intensity, the researchers also recorded reasonably strong þ ratiometric response. The Pr3 ion has several emitting levels in the excited state, and in the (K0.5Na0.5)NbO3 host these levels seem to be differently affected by adsorbed oxygen. Ratiometric sensor material can also be realized in a more straightforward manner by utilizing several differently gas-sensitive constituents. For 3þ instance, a recent work prepared a hybrid material containing Sm :TiO2 nanoparticles (discussed above) attached to the metal-organic framework þ þ Bio-MOF-1 containing Tb3 ions.38 The PL of Sm3 was enhanced, þ whereas that of Tb3 was quenched by oxygen, leading to a strong ratiometric response. Another quite complex composite system based on Sr4Al14O25: þ þ Eu2 ,Dy3 (belonging to a family of persistent phosphors39) was recently studied.40 Samples containing the phosphor particles along with plasmoni- þ cally active silver nanoparticles (in resonance with Eu2 excitation) dispersed in a polymer matrix were prepared in various morphologies. Oxygen þ quenched the Eu2 luminescence up to w4 times, where the oxygen concentration dependence could be described by a modiﬁed SterneVolmer law. The sensing mechanism probably involves autoionization of the excited þ Eu2 ions as an intermediate step, but one should recognize that numerous energy and charge transfer processes have been proposed to describe the PL mechanisms of persistent phosphors in general.39

3D 9.2 Sm :TiO2 9.2.1 Introduction

Titanium dioxide (TiO2) is the most common oxide of titanium with a wide range of applications, including pigments, photocatalysts, and gas sensors. Crystalline TiO2 has two common polymorphs. Bulk crystals of TiO2 usually possess the thermodynamically stable rutile phase, whereas 278 V. Kiisk and Raivo Jaaniso

TiO2 nanocrystals develop mostly in the metastable anatase phase. Nano- crystalline anatase gradually transforms to rutile at temperatures ranging from 600 to 1200C.41 The phase transition is kinetically defined, and the transition temperature depends on crystallite size and purity of the material.41,42 For instance, doping with RE ions results in a higher transition temperature, about 900C for a typical solegel-prepared material.43 The bandgap of anatase is 3.2 eV,44 or a bit higher for nanocrystallites less than 30 nm in diameter.45 This implies that already near-UV light is strongly 3þ absorbed by thin TiO2 films. Moreover, several impurity ions (such as Nd , þ þ þ Sm3 ,Eu3 , and Yb3 ) have been found to emit intense visible or NIR PL 30,46 after being excited through an energy transfer from the TiO2 host. It þ þ was proposed that Sm3 and Nd3 are particularly suitable dopants in anatase TiO2 because their energy levels are located close to the middle of the TiO2 bandgap so that both hole and electron trapping efficiently takes place, whereas autoionization is negligible.30 These features facilitate the use of doped thin films of TiO2 for special luminescence applications such as optical gas sensing, where widespread violet light sources (e.g., light- emitting diodes) can be used for PL excitation combined with compact photodetectors (e.g., photodiodes) for monitoring the PL. 9.2.2 Preparation and characterization of samples 3þ Most of the gas sensing studies of Sm :TiO2 have been carried out on sole e gel-prepared films or powders.25 28,47 The less-pronounced effect was also observed in the case of an atomic layer deposited sample.25 The solegel process is based on the gelation of a colloidal suspension (“sol”) and formation of a continuous inorganic network in the liquid phase (“gel”), as a result of hy- drolysis and polycondensation reactions in a solution containing an organic precursor and water. The organic precursor is typically Ti(OC4H9)4 (titanium butoxide) dissolved in butanol. RE impurity is incorporated by adding proper amount of corresponding salt to the mixture (e.g., SmCl3•6H2O). Gel powder is obtained by dripping the solution to distilled water while stirring. A white precipitate is formed which is dried in an evaporator. RE-related PL of the as-prepared gel is typically very weak due to quenching of the PL by OH-groups and organic residues. The material needs to be heat-treated at temperatures as high as 800C to optimize the PL.48 Up to this temperature, the anatase phase is mostly preserved whereas crystallinity is drastically improved. To reduce agglomeration of the powder particles, the crystallized powder can be dispersed in distilled water using an ultrasound probe. The suspension is then transferred to a glass or silica Rare earthedoped oxide materials for photoluminescence-based gas sensors 279

3þ Figure 9.2 Scanning electron micrograph of solegel-prepared annealed TiO2:Sm nanoparticles deposited on fused silica substrate. Adapted from Eltermann M, Utt K, 3þ Lange S, Jaaniso R. Sm doped TiO2 as optical oxygen sensor material. Opt Mater 2015; 51:24e30. https://doi.org/10.1016/j.optmat.2015.11.020. substrate by drop coating. The resulting material consists of a hierarchical structure of agglomerated nanocrystallites with an average grain size of about 40 nm (Fig. 9.2). BET measurements showed a broad distribution of pore sizes ranging from 25 to 150 nm. The emission spectrum of RE-activated TiO2, when excited with a UV light, generally consists of sharp lines due to the trivalent RE ion as well as a broad emission band peaking at 550 nm (Fig. 9.3). The latter is believed to be intrinsic to anatase-type TiO2, originating either from self-trapped or bound excitons or defect states.44,49 The intrinsic emission becomes quite strong at cryogenic temperatures50 and possibly also under an intense laser excitation as the RE emitter becomes more easily saturated (due to a long lifetime of the excited state). Occasionally, the intrinsic emission itself exhibits rather strong ambient sensitivity,8 but in this series of studies, it was nearly constant and could potentially be used as a reference. 3þ 4 Sm ion has only one major emitting level G5/2 and the four terminal 6 manifolds HJ cause the four emission bands in the visible spectral range. The spectral ﬁne structure is due to the regular crystalline surrounding (crystal- 6 3þ ﬁeld splitting of the HJ states). The intensity of the Sm lines strongly depends on the concentration of oxygen in the ambient environment, but the 280 V. Kiisk and Raivo Jaaniso

4 6 G5/2 H7/2

100 % N2

6 60 % N2 + 40 % O2 H5/2 (a) 6 H9/2

Intensity (a.u.) 6 H11/2

(b)

500 550 600 650 700 750 Wavelength (nm) 3þ Figure 9.3 Photoluminescence emission spectra of the TiO2:Sm nanoparticles excited with 355 nm laser in different ambient environments. Adapted from Elter- 3þ mann M, Utt K, Lange S, Jaaniso R. Sm doped TiO2 as optical oxygen sensor material. Opt Mater 2015;51:24e30. https://doi.org/10.1016/j.optmat.2015.11.020. spectral ﬁne structure and intensity ratio of individual lines is unaffected. It indicates that the crystallographic surrounding of the emitting ions remains unchanged during changes in gas composition, and only one crystalline site 3þ 3þ of Sm is present. Other Sm sites in TiO2 may exist as well, but those are not revealed under indirect excitation.48 9.2.3 PL-based oxygen sensing þ The simplest type of gas sensing response is observed in the Sm3 PL intensity (Fig. 9.4). In each measurement cycle, the sample was exposed to the oxygenenitrogen mixture for 10 min and then to pure nitrogen for 10 min before the next exposure to oxygen. The ﬁnal PL intensity in oxygen-containing ambient monotonically increases with increasing O2 concentration, an exactly opposite behavior to the PL-based sensors described by SterneVolmer law. The PL intensity has certain nonlinear dependence from O2 concentration. The change of the latter is observed over four orders of magnitude (from about 100 ppm trace level up to normal pressure). Moreover, the response is reasonably fast even at room temperature. As shown in the inset of Fig. 9.4, the characteristic response time is under 1 min and recovery time about 5 min. Note that the response time of the material itself should be even smaller, as the measured response includes the instrumental time of changing the gas composition. Rare earthedoped oxide materials for photoluminescence-based gas sensors 281

6 0 I ~1min ~5min / I 5 90 % 90 % 4 2

3 2 2 2 2 0.1 % O 2 0.01 % O Relative PL intensity 100 % O 10 % O 1 1 % O 0 20 40 60 880 1000 Time (mmin)

Figure 9.4 Typical temporal response curve of Sm:TiO2 photoluminescence (PL) intensity for a set of gas exchange cycles covering a wide oxygen concentration range of 100e0.01 vol% (at 25C and normal pressure).28 The strongest Sm3þ emission band around 615 nm has been integrated (Fig. 9.3, with background subtracted).

þ At the same time, a systematic trend was also observed in the Sm3 PL decay kinetics (using a nanosecond pulsed excitation source). The decay became faster as ambient O2 concentration decreased (Fig. 9.5). The effect is quite pronounced and uniform over the wide concentration range so that one can easily employ the PL decay to deﬁne the sensor response independently of the absolute PL intensity. However, the decays strongly diverge

(a) (b) λ 105 λ ex = 355 nm ex = 355 nm λ λ em = 612 nm em = 612 nm 100% O 100% O2 2 4 10 10% 10% 4 1% 10 1% 0.1% 0.1% 0.01% 0.01% 100% N 100% N2 2 103 103 PL intensity (counts) PL PL intensity (counts) PL

102 102 0 100020003000 4000 0 10002000 3000 4000 Time (μs) Time (μs) 3þ 3þ Figure 9.5 Sm photoluminescence (PL) decay kinetics of two TiO2:Sm samples with 3% (A) and 0.5% (B) Sm concentration.28 The colored dots represent experimental data, solid black lines represent theoretical fits (Eq. 9.3), and the dashed straight line represents an exponential decay with characteristic time 300 ms. 282 V. Kiisk and Raivo Jaaniso from a simple exponential decay reflecting the complex nature of the þ excitationeemission route of Sm3 as discussed in the following sections. 9.2.4 Sensing mechanism and its mathematical model Analysis of the PL decay kinetics has turned out to be most informative to understand the PL-based gas sensing mechanism in MOXs. The natural decay of þ Sm3 ions should be about 300 ms as typically detected under direct excitation with visible light.48 Such exponential decay is represented by the dashed lines in Fig. 9.5. The tail of the host-sensitized PL decay is clearly slower and in- þ dicates a delayed excitation of the Sm3 ions (i.e., the excitation energy is trapped in the host matrix for a relatively long time). However, the initial part of the decay is even faster in oxygen-deficient ambient gas, indicating that the gas sensing effect is mostly linked to PL quenching. Hence, it is þ concluded that the excited state of the Sm3 ion (energy donor) is depopu- lated not only by natural decay but also via energy transfer to certain lattice defects (energy acceptors). In turn, the number of such lattice defects is controlled by adsorption processes at the surface. It is plausible that the defects are oxygen vacancies or other intrinsic TiO2 lattice defects, which only in a specific charge state possess excitation energy matching the emission of þ Sm3 . A qualitative scheme of the processes is depicted in Fig. 9.6. A corresponding mathematical model is built as follows. Assuming that the acceptors are randomly distributed in the material, the PL decay should follow the well-known law51 ð = Þð Þb uðtÞ¼e k0t c c0 k0t (9.1)

Here k0 is the rate constant for natural decay ( ¼ 1=300 ms). The power index b depends on the type of interaction between the donors and acceptors and on the Euclidean dimensionality of the acceptor space. The most common case is dipolar interaction in 3D spacedthe Forster€ ’s energy transfer with b ¼ 1=2. Parameter c is the acceptor concentration, while the constant c0 characterizes the energy transfer strength and is inversely proportional to the third power of the characteristic (Forster€ ’s) energy transfer radius. þ The delayed excitation of Sm3 ions can be attributed to trapping of the initial excitation (polaron or some kind of exciton) in the TiO2 host for an þ extended period before the energy is transferred to a nearby unexcited Sm3 ion. The binding energy of these trap states must be relatively small (comparable with the thermal vibration energy kBT), as the resulting escape times are still as small as several milliseconds at room temperature. Indeed, several Rare earthedoped oxide materials for photoluminescence-based gas sensors 283

Conduction band – O2 ab

UV Shallow Quenching excitation traps defect Crystallite surface PL

PL center (Sm3+ ion)

Valence band

Radiative processes Non-radiative processes Energy transfer Surface electron trapping 3þ 3þ Figure 9.6 A simpliﬁed energy diagram of the TiO2:Sm sensor material. The Sm ion is exited either instantly (blue dashed line) or in a delayed manner (red dashed line). Once the excitation has reached the Sm3þ ion, it can (a) emit a photon or (b) be quenched by a defect. This defect can be “switched off” by electron transfer to surface oxygen species. PL, photoluminescence. Adapted from Eltermann M, Kiisk V, Berholts A, Dolgov L, Lange S, Utt K, Jaaniso R. Modeling of luminescence-based oxygen sensing by redox- 3þ switched energy transfer in nanocrystalline TiO2:Sm . Sensor Actuator B Chem 2018; 265:556e64.

52e54 studies suggest the existence of such shallow traps in TiO2. Assuming that the delayed population occurs at a constant rate k (corresponding to the presence of a single type of trap level), the decay law (1) can be generalized as follows: kt IðtÞ¼I0uðtÞþI1ke 5uðtÞ; (9.2) where the ﬁrst term describes the instantly excited PL centers and the second term those PL centers excited with a delay. The 5 symbol marks convolution. The convolution reﬂects the fact that different PL centers start to decay at different time delays from the laser pulse, and the convolution sums over all possible delay times from 0 to t. It was found that the experimental decay curves were accurately reproduced only by assuming the presence of traps with different depths, leading to a distribution rðkÞ of delayed population rates. Eq. (9.2) is now further generalized to the equation ZN kt IðtÞ¼I0uðtÞþI1 dk rðkÞke 5uðtÞ; (9.3) 0 284 V. Kiisk and Raivo Jaaniso

Presumably, the result is not very sensitive to the choice of rðkÞ. Good fitting of the decays (solid lines in Fig. 9.5) can be obtained by assuming thermally activated release of carriers from traps with an exponential distribution of trap depths.28 The parameter b may also require adjustment for þ high Sm3 concentrations (where cross relaxation becomes dominant), segregation of the impurity (due to annealing at high temperatures55), or rather small radius of the crystallites (where the acceptor space cannot be considered infinite). One can now study the effect of oxygen adsorption on the different model parameters derived from the fitting. It was found that only the relative acceptor concentration c =c0 changed substantially and decreased systematically with increasing O2 content in the ambient gas (Fig. 9.7). So the concentration change of the acceptor centers seems to be the main factor 3þ connecting the changes in O2 concentration and PL intensity of Sm . Note that within the frames of the proposed model, the delayed excitation complicates only the decay function (Eq. 9.3) and increases the average lifetime but does not alter the stationary PL signal and its dependence on the oxygen pressure. This is because each excitation, which is initially trapped in þ the host, will finally end up exciting a Sm3 ion. In the case of Forster€ ’s energy transfer, one can explicitly derive the stationary PL intensity (i.e., area under the decay curve)28: I þ I pffiffiffi À Á Ã S ¼ 0 1 1 p q exp q2 erfcðqÞ ; k0 where 2q ¼ c=c0. As one can see, the equation does not include any details on the traps causing the delayed PL and the intensities I0 and I1 are simply added. Further elaboration of the model requires assumptions on O2 adsorption and related processes. Adsorption of gaseous O2 on MOX surfaces is generally accompanied by a charge transfer and formation of various negatively e charged oxygen species at the surface (O2 being dominant at room temper- ature56,57). The electron involved in the charge transfer could be taken from the acceptor defects (either directly or through the conduction band of TiO2). Different adsorption mechanisms exist depending on the defective state and coverage of the surface (including the presence of hydroxyl groups).58,59 Therefore, the adsorption isotherm can be quite complex. It 28 was found that the changes in the derived acceptor concentration c =c0 Rare earthedoped oxide materials for photoluminescence-based gas sensors 285

9.0

101 EPL 8.5 S

8.0 –13.5

–14.0 EC ](%) 2 S

–14.5 [O

100

9.3 IPL

S 9.2

9.1 0 1020304050 Time (h) 3þ Figure 9.7 The temporal behavior of extrinsic (Sm ) photoluminescence (PL) (SEPL), intrinsic PL (SIPL), and electrical conductivity (SEC) of the TiO2 nanopowder, while being excited by 365 nm LED and subjected to a randomly changing oxygen concentration.7 All signals are represented as logarithms, whereas data point color encodes the actual oxygen concentration varied between 0.21% and 21%. The experiment was done at room temperature.

can be connected to the changes of O2 surface density, assuming that the latter follows the Toth isotherm: ð Þm 1=m cads ¼ Kx ; m csat 1 þ ðKxÞ where x is gaseous O2 concentration (or partial pressure) and K; m are parameters of the isotherm. This isotherm is a generalization of the Langmuir and Freundlich isotherms and is believed to be more suitable for a broad pressure range and heterogeneous substrates. We note that photoadsorption and -desorption of O2 on clean or hydrox- 57,60,61 ylated TiO2 surfaces has also been known for a long time and may determine some essential aspects of the gas sensing, such as response time or drift of the PL signal. Impact of both photochemistry and humidity must be taken into account in the future studies and applications of this sensor. 286 V. Kiisk and Raivo Jaaniso

3þ 9.2.5 Multivariable sensing with TiO2:Sm

The photoexcited charge carriers in TiO2 are known to cause quite significant photoconductivity, which also responds to the ambient environment.62 It would be intriguing to register simultaneously the correlated responses of photoconductivity and PL, to improve selectivity or accuracy of the sensor and gain further insight into the complex energy and charge transfer processes. Indeed, a recent study7 showed that the electrical conductance of a thin layer of solegel-prepared TiO2:Sm powder (similar to the one discussed above) had almost one-to-one correspondence to the lumi- þ nescence of Sm3 ions, while the ambient oxygen concentration (in a ﬂow- ing O2/N2 mixture) was randomly varied (Fig. 9.7). Considering the anticipated conduction mechanism (through the double-Schottky barriers formed between contacting nanoparticles), it is likely that adsorption or desorption of oxygen causes not only recharging of the acceptor defects þ (quenching Sm3 PL) but also changes the extent of the space charge layers on crystallite surfaces, leading to a respective change in electrical conductance. In the cited study, the oxygen sensitivity of the intrinsic broadband emission (with a spectrum similar to the one shown in Fig. 9.3) was also recorded (Fig. 9.7). The response of the intrinsic PL is much slower and has possibly a different mechanism.8 Nevertheless, it can deliver complementary information as a sensor signal. Figure 9.8 depicts the pairwise correlations between the signals. Interest- ingly, there are almost no occurrences where different O2 concentrations

[O2](%) 100 101 EC IPL IPL S S S

S S S EPL EC EPL Figure 9.8 The three signals from Fig. 9.7 plotted pairwise against each other.7 Rare earthedoped oxide materials for photoluminescence-based gas sensors 287 map to the same coordinates in these two-dimensional spaces. The þ space spanned by the Sm3 PL intensity and photoconductivity is quite squeezed. However, the defect-related PL intensity in combination with one of the remaining two signals covers a wide space and could potentially more reliably predict the ambient O2 concentration and suppress the sensor drift (which is clearly present in the individual signals shown in Fig. 9.7). Mapping combinations of the three signals (features) to corresponding O2 concentrations (target) is a typical machine learning problem. For instance, if two sensor signals S1 and S2 are involved, then the predicted oxygen concentration p½O2¼f ðS1; S2Þ, where f is a convenient mathematical function involving a number of free parameters that can be “trained” to fit the data. Because of the rather smooth relationships between lnðp½O2Þ and the logarithms of the measured signals (Fig. 9.8), it is justified to consider lnðp½O2Þ as a function of S1 and S2 and expand the unknown function into a Taylor series: ð ½ Þ ¼ þ þ þ 2 þ þ 2 þ . ln p O2 a0 a1S1 a2S2 a11S1 a12S1S2 a22S2 (9.4) The coefficients a0, ai, ai;j, etc. (which represent the partial derivatives) can be considered as the free parameters of the model. They can be determined (trained) by minimizing a cost function taken as the residual sum of squared errors between the logarithms of the predicted and the corresponding true O2 concentrations. The advantage of the approach is that due to the linear dependence of the model on the parameters and the particular choice of the cost function, the optimization is an ordinary least square (OLS) problem. Regardless of the number of variables, OLS leads to a system of linear equations for the optimal parameter values, so that one can directly find the global optimum without a complex and time-consuming iterative process required for training a neural network. Models with different orders and signal combinations were tested, where the initial 32 h (60%) of data was used for training and the remaining data points were used to evaluate the mean relative error for the model prediction. It turned out that in most cases a linear relation (first-order polyno- mials) provided the best results. As an example, Fig. 9.9 demonstrates how fusing electrical conductance and intrinsic PL yield significant improvement of precision in predicting oxygen concentration. More than 4 times of reduction in the relative error was achieved. 288 V. Kiisk and Raivo Jaaniso

3D 9.3 Eu :ZrO2 9.3.1 Introduction

Zirconium dioxide (ZrO2) is a well-known refractory material primarily used in hard ceramics, gemstones, abrasives, and optical coatings. At ambient conditions, the stable phase of nominally pure zirconia is the monoclinic one (m-ZrO2, naturally found as the mineral baddeleyite). At high temperatures, a martensitic transformation ﬁrst into tetragonal (t- 63 ZrO2) and then into cubic (c-ZrO2) polymorph takes place. Substituting þ þ þ Zr4 with aliovalent impurities (such as Ca2 or Y3 ) induces charge- compensating defects (such as anion vacancies), which due their structural distortion can stabilize the metastable t-ZrO2 or c-ZrO2 phases at room temperature.64 Especially the cubic phase possesses a remarkable ionic conductivity facilitating its use in fuel cells and high-temperature oxygen sensors. Tetragonal phase may also develop in a nanozirconia (e.g., powders with crystallite size less than about 30 nm) as a result of excess surface or strain energies.65,66 67 The bandgap of m-ZrO2 is around 5.7 eV, which corresponds to optical transparency down to 220 nm. Nevertheless, even photons with longer wavelength (w280 nm) can excite certain intense broadband PL, usually attributed to oxygen vacancies or residual impurities.68 This broadband e PL sometimes shows an oxygen sensitivity,69 71 but the effect is somewhat

1.0 EC EC+IPL 0.5

0.0

Relative error –0.5

–1.0 0 1020304050 Time (h) Figure 9.9 The relative errors of two different calibration functions obtained through ordinary least squares optimization.7 Model “EC” involves only photoconductivity, whereas “ECþIPL” combines conductivity and intrinsic PL using Eq. (9.4). Only the points to the left of the red line were used for training, whereas the remaining data points served for testing the model. In this experiment, the oxygen concentration was varied between 0.21% and 2.1%. Rare earthedoped oxide materials for photoluminescence-based gas sensors 289 controversial, possibly because several different PL centers contribute to the wide emission spectrum. Moreover, this PL requires a deep UV excitation and exhibits a strong temperature dependence. However, the wide bandgap can accommodate various optical impurities. Because of charge difference (and sometimes also size mismatch), trivalent RE ions do not naturally substitute into MeO2-type crystalline hosts (compared with hosts like Y2O3). The large amount of charge- compensating defects and the phase stability issues may limit the performance and maximum useable concentration of RE ions for luminescence 3þ applications. At least in the case of ZrO2:Er , it has been reported that niobium (Nb) codopant improves the (upconverted) luminescence performance as well as stability of the dominant monoclinic phase.72,73 It was 5þ shown that in ZrO2:Eu,Nb, the niobium was incorporated as Nb , at least close to the surface.34 Hence, it is believed that a comparable amount of Nb þ þ codopant compensates the charge difference of the RE3 and Zr4 ions. þ þ þ Moreover, the ionic radii of Zr4 ,Eu3 , and Nb5 are 78, 101, and 69 pm, respectively (due to Shannon,74 for coordination number 7, as in m- 5þ ZrO2). Hence, there is a chance that Nb also compensates for the lattice þ distortion induced by the RE3 ion. 9.3.2 Preparation and characterization of samples Zirconia nanopowders can be prepared by using various solegel routes. The particular materials used for the gas sensing experiments were prepared by using solegel combustion technique, where glycine was used as fuel and 72,75 nitric acid as the oxidant. ZrCl4 was dissolved in methanol, whereas Nb2O5 and Eu2O3 were dissolved in nitric acid. Appropriate amounts of the MOX solutions and glycine were mixed. The resulting mixture was evaporated on a hot plate while stirring at 90e100C and concentrated until a gel consistence was obtained. The gel was heated to 300e350Cinan open oven for 2 h to promote combustion to eliminate the nitric oxide. The obtained black powder was ﬁnally annealed at 1200C for 2 h resulting in a white powdered material. Scanning electron microscopy showed strongly agglomerated particle-like formations with diameters ranging from 200 to 600 nm (Fig. 9.10). Only a submicron porous network is resolved. X-ray diffraction (XRD) and Raman scattering analyses showed conclu- sively the impact of impurity content on the crystal structure. At low impurity concentrations, the material was mostly monoclinic although nonnegligible amount (3e6 at%) of the tetragonal phase was also present. 290 V. Kiisk and Raivo Jaaniso

Figure 9.10 Scanning electron microscopy micrographs of the solegel-derived ZrO2: Eu,Nb powder (annealed at 1200C) at two different magniﬁcations.34

By contrast, 8at% of europium impurity could stabilize the material in the tetragonal phase, leaving no traces of m-ZrO2. However, a similar amount of Nb codopant strongly suppressed the formation of t-ZrO2 so that m-ZrO2 became again dominant. XRD analysis also established crystallite sizes of about 50 and 25 nm in the monoclinic and stabilized tetragonal phases, respectively. Lower heat treatment temperatures can give smaller crystallites (which can be advantageous for gas sensing), but their phase purity and luminescence performance are compromised. 3þ Owing to the large bandgap of ZrO2,thePLofEu can be excited þ either directly, through the O2 /Eu3 charge transfer absorption band (around 250 nm for m-ZrO2), or the energy transfer after band-to-band 76,77 3þ excitation. There is one prevalent Eu site in both m-ZrO2 and t-ZrO2 with different symmetry properties, resulting in a clear distinction of the spectral ﬁne structures. In the oxygen sensing studies, direct excita- 5 7 tion at 395 nm was used (intra-4f transition D0/ FJ). The obtained PL þ spectra of Eu3 correlate well to the dominant crystal structure (Fig. 9.11). 9.3.3 Oxygen sensing 3þ 3þ Similarly to TiO2:Sm , the PL of ZrO2:Eu was also found to be sensitive to ambient oxygen.34,35 However, the temperature, the content of Nb codopant, annealing conditions, and even excitation laser intensity all inﬂu- enced the size and sign of the response. Most of the oxygen response studies were conducted at 300C because the response was stronger at elevated temperatures. Rare earthedoped oxide materials for photoluminescence-based gas sensors 291

λexc=395 nm T=23°C

Eu 2

Eu 8 PL intensity (arb. units) PL

Eu 8.68 Nb 8.12

580 590 600 610 620 630 640 650 660 Wavelength (nm)

Figure 9.11 Photoluminescence (PL) emission spectra of solegel-derived ZrO2:Eu and ZrO2:Eu,Nb powders annealed at 1200 C. The concentration of dopants (in at%) is shown by the numbers after Eu and Nb. Adapted from Kiisk V, Puust L, Mandar€ H, Ritslaid P, Bite I, Jankovica D, Sildos I, Jaaniso R. Phase stability and oxygen-sensitive photoluminescence of ZrO2:Eu,Nb nanopowders. Mater Chem Phys 2018;214:135e42. All samples systematically responded to the change in oxygen concentration (Fig. 9.12). The best performance was demonstrated by the sample containing 1.48at% Eu, which was overcompensated by Nb (2.74at%). In this case, the PL intensity became stronger as oxygen concentration increased. By contrast, the responses of the materials containing only Eu (either 2 or 8at%) were reversed. Usually the stable behavior shown in Fig. 9.12 was recorded only during a repeated cycle of gas exposures. Fig. 9.13 shows an extended measurement. At the beginning, one can recognize certain slow background process with a characteristic time constant w100 min, which affects the absolute PL intensity. The signal becomes stable after one to two cycles. 9.3.4 Sensing mechanism Several attempts were made to identify the sensing mechanism from the PL decay kinetics, using either pulsed or modulated laser for excitation. At least in the case of ZrO2:Eu(1.48at),Nb(2.74at%), it is quite certain that quench- þ ing of Eu3 ﬂuorescence by random acceptors are involved, similarly to 292 V. Kiisk and Raivo Jaaniso

1,0

0,8

0,6

0,4 Eu (2) 1,0

0,8 λexc=395 nm 0,6 T=300°C 0,4 Eu (2.12) Nb (1.87) 1,0 Eu (1.48) Nb (2.74) 0,8

0,6

Intensity 0,4

1,0 2 2 2 2 2 0,8 2 2 2 2 2 2

0,6 6%O 50%O 25%O 12%O 2 2 2 2 100% N 100% N 100% N 100% N 100% N 100% N 100% O 0,4

94% N Eu (8) 1,0 50% N 75% N 88% N

0,8

0,6

0,4 Eu (8.68) Nb (8.12)

0 102030405060708090100110120 Time (min) 3þ Figure 9.12 Temporal behavior of the Eu PL intensity of the ZrO2:Eu and ZrO2:Eu,Nb powders in response to the ambient oxygen concentration changes at 300C.35

3þ 3þ TiO2:Sm .AsEu is directly excited, no delayed luminescence is observed in this case. The effect of changing the atmosphere was quite pronounced in the case of pulsed excitation (see the curves taken at 300Cin Fig. 9.14). In an oxygen enriched ambient, nearly perfect single exponential þ decay of Eu3 was observed with a time constant about 1.1 ms. This can be 5 attributed to the natural lifetime of the D0 excited state. tblt fteresponse. the of stability ifrn min conditions. ambient different h aenme fectto pulses. excitation of number same the xgnexposure. oxygen atccecnan lentn xoue o10 O 100% to exposures alternating contains cycle last iue9.13 Figure earth Rare iue9.14 Figure e oe xd aeil o htlmnsec-ae a sensors gas photoluminescence-based for materials oxide doped

PL intensity (arb. u.) ZrO of behavior Typical Eu 10 10 10 10 10 3

þ Normalized intensity –3 –2 –1 0 1 35 0.0 0.2 0.4 0.6 0.8 1.0 Ldcykntc fteZrO the of kinetics decay PL 0040 0080 00 20 14000 12000 10000 8000 6000 4000 2000 0 The 0 ﬁ s w ylsaeietclt hs in those to identical are cycles two rst 0101020203030400 350 300 250 200 150 100 50 34 h neste ftedcycre r omlzdto normalized are curves decay the of intensities The λ λ det exc 2 =615 nm :Eu =464 nm 3 þ Time (min) Time ( Litniydrn he ucsiecce of cycles successive three during intensity PL μ s) 2 E(.8t,b27a% odrin powder :Eu(1.48at),Nb(2.74at%) 2 n 0%N 100% and N N O O 2 2 2 2 (300 (23 (23 (300 i.9.12 Fig. ° ° C) C) ° ° C) C) 2 ots o the for test to hra the whereas 293 294 V. Kiisk and Raivo Jaaniso

Eu (2) Oxygen Nitrogen

0,0 0,1 0,2 Nitrogen

Oxygen Eu (8.68) Nb (8.12) 1000°C PL intensity (arb. units) PL 01234567 01234567 Time (ms) Time (ms) 3þ Figure 9.15 Eu PL decay kinetics of a ZrO2:Eu and ZrO2:Eu,Nb powder in 100% O2 and 100% N2, demonstrating different sensing mechanisms as well as the reversal of the response. Adapted from Kiisk V, Puust L, Mandar€ H, Ritslaid P, Bite I, Jankovica D,

Sildos I, Jaaniso R. Phase stability and oxygen-sensitive photoluminescence of ZrO2:Eu,Nb nanopowders. Mater Chem Phys 2018;214:135e42.

For ZrO2:Eu,Nb materials with other impurity concentrations or annealing conditions, the interpretation is less clear. Although the reported PL decays are qualitatively in agreement with the corresponding stationary PL responses, several different types of kinetics were observed, of which a few distinct variants are depicted in Fig. 9.15. The more strongly curved PL decays of the uncompensated ZrO2:Eu(2at%) are in agreement with the assumption that oxygen vacancies behave as PL quenching centers, yet the gas response is unexpectedly reversed, compared with ZrO2:Eu(1.48at %),Nb(2.74at%). Even more interestingly, quite a different sensor response þ was observed at high doping levels, where some Eu3 emitters are effectively switched on or off by the change of ambient gas, resulting in the apparent vertical shift of the PL kinetics. However, it is unlikely that one phosphor material can show several completely unrelated gas sensing mechanisms. Assuming that the PL quenching centers and electron donors are oxygen vacancies, there might be an interplay between vacancies in different charge þ states and how these vacancies are positioned with respect to Eu3 sites.

3D 9.4 Tb :CePO4 9.4.1 Introduction

Orthophosphates, such as LnPO4 (where Ln represents a trivalent lanthanide ion), constitute another popular class of host materials for RE emitters. RE-rich LnPO4 occurs naturally as varieties of the mineral monazite possessing a monoclinic structure. Rhabdophane is a hydrated, hexagonal form of CePO4, transforming irreversibly to the monazite phase after heating at about 800C.78,79 Monazite is thermally stable.80 Rare earthedoped oxide materials for photoluminescence-based gas sensors 295

þ In practical phosphors, Ce3 is almost always used as a sensitizer for 3þ 81 3þ 3þ 3þ Tb . However, in nanocrystalline LnPO4:Ce ,Tb or CePO4:Tb þ þ phosphors, Ce3 is easily oxidized to Ce4 , unless the synthesis is specially þ þ elaborated to protect Ce3 .82,83 Ce4 quenches the luminescence of þ þ þ Tb3 , whereas energy transfer from Ce4 to Tb3 is impossible.82,84 Such kind of disadvantages for conventional phosphors can be favorable for sensor applications. 9.4.2 Preparation and characterization of samples

Differently from TiO2:Sm and ZrO2:Eu, the gas sensing experiments with 3þ CePO4:Tb were conducted on a material consisting of rather fine and regularly shaped nanorods.36 The material was prepared by a simple aqueous route based on the standard Schlenk technique, using CeCl3, TbCl3, and H3PO4 as precursors (note that there are also other aqueous routes resulting 79 in LnPO4 nanorods or -wires ). The as-prepared product was annealed at 300C in a reducing atmosphere for 2 h to remove the structural water. The þ concentration of Tb was 10at%, where presumably Tb3 substitutes for þ Ce3 ions in the crystal lattice. The synthesis resulted in single-crystalline (rhabdophane-type) nanorods with a quite regular shape, having a width of 10e20 nm. The size and morphology resulted in a large surface-to-volume ratio with a specific surface area of 176 m2 g 1. þ The 4f/5d electronic transitions of Ce3 result in several overlapping absorption peaks in the deep UV region. In the particular material, most efficient excitation of PL occurred over 250e300 nm (Fig. 9.16). The broad PL þ band of Ce3 ion is also located in the UV region, but is quite weak, as most þ of the excitation energy is transferred to Tb3 ions emitting green light þ (quantum yield as high as 50% for the Tb3 emission was reported). 9.4.3 Gas sensing and its mechanism Gas sensing experiments were conducted under quite aggressive conditions. At 200C, the material was exposed alternately to highly oxidizing (100% oxygen) or reducing (95% N2/5% H2) atmospheres. When exposed to oxygen, the PL consistently decreased to a negligible level with a characteristic response time of a few minutes (Fig. 9.17). When exposed to 95% N2/5% H2, the PL recovered in a similar manner. For comparison, microcrystals of CePO4:Tb prepared by solid-state reaction were also measured, and even such bulk material showed a measurable PL response (although the effect was very small, w5%). These observations suggest that, due to the small 296 V. Kiisk and Raivo Jaaniso

5 7 D4- F5

200 250 300 350 400 Wavelength / nm Intensity / a.u. 5 7 5D -7F D4- F6 4 4 5 7 D4- F3 Ce(III)

300 350 400 450 500 550 600 650 Wavelength / nm 3þ Figure 9.16 Photoluminescence (PL) emission spectrum of CePO4:Tb nanorods. PL excitation spectrum is shown in the inset. Adapted from Di W, Wang X, Xinguang R.

Nanocrystalline CePO4:Tb as a novel oxygen sensing material on the basis of its redox responsive reversible luminescence. Nanotechnol 2010;21:075709.

100

20 Normalized intensity / a.u. O2 N2+H2

0 0 20 40 60 80 100 Time / min 3þ Figure 9.17 Development of the Tb PL intensity of nanocrystalline CePO4:Tb when alternately subjected to 100% O2 and 95% N2/5% H2 atmospheres. Adapted from Di W, Wang X, Xinguang R. Nanocrystalline CePO4 :Tb as a novel oxygen sensing material on the basis of its redox responsive reversible luminescence. Nanotechnol 2010;21:075709. Rare earthedoped oxide materials for photoluminescence-based gas sensors 297

0.1 a b c d

0.01 Intensity (a.u.)

1E–3 0.000 0.005 0.010 0.015 0.020 0.025 Time (s) 5 Figure 9.18 Photoluminescence decay kinetics of CePO4:Tb nanocrystals (from the D4 level of Tb3þ) for the original sample (a) and those exposed to 40% (b), 80% (c), and 100% (d) oxygen atmospheres at 200C for 5 min, respectively. Solid curves describe exponential (a) or biexponential ﬁts (aec). Adapted from Di W, Wang X, Xinguang R.

Nanocrystalline CePO4 :Tb as a novel oxygen sensing material on the basis of its redox responsive reversible luminescence. Nanotechnol 2010;21:075709. size of the nanocrystals, the chemical change induced by the ambient gas has þ an exhaustive effect on the Tb3 centers, resulting in the very high contrast observed in the PL intensity. In this case, the chemical effect of the ambient gas was explicitly identi- ﬁed using X-ray photoelectron spectroscopy (XPS). XPS spectrum of the sample exposed to oxygen showed clearly an additional peak characteristic þ þ of Ce4 ion. The peaks due to Ce3 were diminished but remained quite þ þ strong. Hence, the reduced energy transfer from Ce3 to Tb3 cannot fully explain the drastic suppression of the PL intensity. Measurement of PL decay þ kinetics showed that the PL of Tb3 was additionally quenched (Fig. 9.18), þ probably by the created Ce4 centers. All described effects were found to be reversible. This gas sensing mechanism assumes direct reaction between þ þ Ce3 /4 ions and gas molecules and is therefore limited to rather small nanocrystals. Again, the sensor response can be deﬁned based on either PL intensity or PL decay time. For this material, the dependence on oxygen concentration was more linear (compared with TiO2:Sm and ZrO2:Eu), making such an optical sensor appropriate for operation at high oxygen concentrations. 298 V. Kiisk and Raivo Jaaniso

3D 9.5 Pr :(K0.5Na0.5)NbO3 9.5.1 Introduction 3þ The recently reported oxygen-sensitive PL of Pr :(K0.5Na0.5) 37 NbO3 represents an interesting case. First, the material consists of microcrystals (rather than nanocrystals). Second, the activator was praseodymium þ ion (Pr3 ), which is one of the few RE ions exhibiting several emitting levels. Therefore, the potentially dual optical response can be obtained. It is common to utilize such emitters for optical temperature sensing where populations of the emitting levels are in thermal equilibrium (this has also been realized for the particular material85), but the approach is novel in the context of gas sensors. The host, potassiumesodium niobate, is otherwise known as a promising lead-free piezoelectric ceramic.86 The optical bandgap of the bulk material with perovskite (orthorhombic) structure has been estimated to be w4.3 eV.87 Some synthesis routes result in bandgap values as small as 3.2 eV.88 3þ 3 The main levels of Pr -producing emission in the visible range are P0 1 and D2. The latter gives an emission band at w600 nm due to the 1 3 3 D2/ H4 transition. A series of emission bands originating from the P0 level are observed in the visible range, the most prominent at w500 nm be- 3 3 ing due to the P0/ H4 transition. Ratio of the emissions depends on excited state dynamics and is determined by the host, excitation route, and concentration of the activators.89 9.5.2 Synthesis þ The samples with 0.5at% of Pr3 were fabricated via a conventional solid- state reaction, which is commonly used for the synthesis of various phosphors. K2CO3,Na2CO3,Nb2O5, and Pr6O11 were ball-milled with the addition of alcohol and then calcined in an alumina crucible at 880C. The obtained material was remilled, mixed with polyvinyl alcohol, and pressed into pellets, which were ﬁnally sintered at 1100C. As a result, (K0.5Na0.5)NbO3 microcrystals (1e4 mm in size) with the orthorhombic crystal structure and cubic morphology were obtained. The pellets were additionally annealed either in argon or oxygen at 950C. 9.5.3 Oxygen sensing

Using excitation at 325 nm, the PL intensity in 100% N2 and 100% O2 atmospheres at 1 bar was compared.37 Notable change of the PL intensity Rare earthedoped oxide materials for photoluminescence-based gas sensors 299 was observed only at elevated temperatures, using the material annealed 3 3 1 3 in argon. At 98 C, both P0/ H4 and D2/ H4 emissions were present, but only the latter signiﬁcantly (more than 2) responded to the switching 1 3 of ambient atmosphere. At 165 C, the response of the D2/ H4 emission 3 3 became even stronger (more than 3), whereas the P0/ H4 emission was quenched and probably unusable. Therefore, several detection protocols are feasible with this material, including dual and ratiometric response. 1 3 At 165 C, the D2/ H4 emission systematically responded to oxygen concentration down to 2%. At high oxygen concentrations, the response was quite quick (about 1e2 min), but slowed down at lower concentra- 3þ tions. The overall behavior resembles that of TiO2:Sm (see Section 9.2). Although PL decay kinetics were not recorded in this case, the presented data are compatible with the mechanism involving resonant energy transfer to certain defects affected by oxygen adsorption. In particular, the high relative response achieved with micron-sized crystals indicates that the effective interaction range is quite large. The thickness of the electron depletion layer resulting from oxygen adsorption is either comparable to the penetration depth of the 325 nm excitation light or extends throughout the crystal. 3 3 The weak response of the P0/ H4 emission may indicate that the transition is not in a good resonance with the energy acceptor (defect). Never- theless, the authors propose the involvement of a secondary sensing mechanism, where oxygen adsorption (through electron trapping) affects the position of the intervalence charge transfer state.

9.6 Conclusion The existing results have convincingly demonstrated that the luminescence of trivalent RE ions doped into MOX nano- or even microcrystals can exhibit a pronounced and systematic response to an ambient gas, where the sensing mechanism is fundamentally different from the SterneVolmer type collisional quenching of organic fluorophores. The mechanism is usually based on a fluorescence quenching process of the RE emitter coupled to a gas adsorptionerelated redox process. In some cases, both excitation and emission routes are affected. Because of this multistage sensing mechanism, the gas concentration dependence of the response is inherently more complex (albeit in many cases empirically very close to a power law). The response of several oxygen- 3þ 3þ sensitive nanophosphors (TiO2:Sm , ZrO2:Eu ) is such a function of 300 V. Kiisk and Raivo Jaaniso oxygen concentration which allows reaching the trace levels (100 ppm) and spanning a wide dynamic range. Real-world applications necessitate further studies to evaluate and improve the stability, accuracy, and specificity (including reducing the influence of humidity) and diversify the response (i.e., also detect reactive gases other than oxygen). Some advancement may stem from proper engineering of surface morphology and functionalization. On the other hand, materials 3þ 3þ containing multiple RE emitters (such as TiO2:Sm ,Nd ) or RE ions þ exhibiting several emitting levels (such as Pr3 ) are potentially capable of the dual optical response. Moreover, intrinsic luminescence and (photo) conductivity can complement RE luminescence, providing multivariable output from a single sensor. References 1. Wang X, Wolfbeis OS. Optical methods for sensing and imaging oxygen: materials, spectroscopies and applications. Chem Soc Rev 2014;43:3666e761. 2. Gregory JW, Sakaue H, Liu T, Sullivan JP. Fast pressure-sensitive paint for flow and acoustic diagnostics. Annu Rev Fluid Mech 2014;46:303e30. https://doi.org/10.1146/ annurev-fluid-010313-141304. 3. Hodgkinson J, Ralph PT. Optical gas sensing: a review. Meas Sci Technol 2013;24: 012004. 4. Pallotti DK, Passoni L, Maddalena P, Di Fonzo F, Lettieri S. Photoluminescence mechanisms in anatase and rutile TiO2. J Phys Chem C 2017;121:9011e21. https://doi.org/ 10.1021/acs.jpcc.7b00321. 5. Cho B, Hahm MG, Choi M, Yoon J, Kim AR, Lee Y-J, Park S-G, Kwon J-D, Kim CS, Song M, Jeong Y, Nam K-S, Lee S, Yoo TJ, Kang CG, Lee BH, Ko HC, Ajayan PM, Kim D-H. Charge-transfer-based gas sensing using atomic-layer MoS2. Sci Rep 2015;5: 8052. 6. Zhyrovetsky VM, Popovych DI, Savka SS, Serednytski AS. Nanopowder metal oxide for photoluminescent gas sensing. Nanoscale Res Lett 2017;12:132. https://doi.org/ 10.1186/s11671-017-1891-5. 7. Eltermann M, Kiisk V, Lange S, Jaaniso R. Multivariable oxygen sensing based on photoconductivity and photoluminescence of TiO2 nanoparticles. Sensor Actuator B Chem 2019 (in print). 8. Pallotti DK, Passoni L, Gesuele F, Maddalena P, Di Fonzo F, Lettieri S. Giant O2- induced photoluminescence modulation in hierarchical titanium dioxide nanostructures. ACS Sens 2017;2:61e8. https://doi.org/10.1021/acssensors.6b00432. 9. Valerini D, Cretì A, Caricato AP, Lomascolo M, Rella R, Martino M. Optical gas sensing through nanostructured ZnO films with different morphologies. Sensor Actuator B Chem 2010;145:167e73. https://doi.org/10.1016/j.snb.2009.11.064. 10. Sanchez-Valencia JR, Alcaire M, Romero-Gomez P, Macias-Montero M, Aparicio FJ, Borras A, Gonzalez-Elipe AR, Barranco A. Oxygen optical sensing in gas and liquids with nanostructured ZnO thin films based on exciton emission detection. J Phys Chem C 2014;118:9852e9. https://doi.org/10.1021/jp5026027. 11. Faglia G, Baratto C, Sberveglieri G, Zha M, Zappettini A. Adsorption effects of NO2 at ppm level on visible photoluminescence response of SnO2 nanobelts. Appl Phys Lett 2005;86:011923. https://doi.org/10.1063/1.1849832. Rare earthedoped oxide materials for photoluminescence-based gas sensors 301

12. Setaro A, Bismuto A, Lettieri S, Maddalena P, Comini E, Bianchi S, Baratto C, Sberveglieri G. Optical sensing of NO2 in tin oxide nanowires at sub-ppm level. Sensor Actuator B Chem 2008;130:391e5. https://doi.org/10.1016/j.snb.2007.09.015. 13. Liu X, Du B, Sun Y, Yu M, Yin Y, Tang W, Chen C, Sun L, Yang B, Cao W, Ashfold MNR. Sensitive room temperature photoluminescence-based sensing of H2S with novel CuOeZnO nanorods. ACS Appl Mater Interfaces 2016;8:16379e85. https://doi.org/10.1021/acsami.6b02455. 14. Comini E, Baratto C, Faglia G, Ferroni M, Sberveglieri G. Single crystal ZnO nanowires as optical and conductometric chemical sensor. J Phys D Appl Phys 2007;40:7255. 15. Aad R, Simic V, Cunff LL, Rocha L, Sallet V, Sartel C, Lusson A, Couteau C, Lerondel G. ZnO nanowires as effective luminescent sensing materials for nitroaromatic derivatives. Nanoscale 2013;5:9176e80. https://doi.org/10.1039/C3NR02416D. 16. Bismuto A, Lettieri S, Maddalena P, Baratto C, Comini E, Faglia G, Sberveglieri G, Zanotti L. Room-temperature gas sensing based on visible photoluminescence properties of metal oxide nanobelts. J Opt A Pure Appl Opt 2006;8:S585e8. https://doi.org/ 10.1088/1464-4258/8/7/S45. 17. Liu X, Sun Y, Yu M, Yin Y, Yang B, Cao W, Ashfold MNR. Incident fluence dependent morphologies, photoluminescence and optical oxygen sensing properties of ZnO nanorods grown by pulsed laser deposition. J Mater Chem C 2015;3:2557e62. https:// doi.org/10.1039/C4TC02924K. 18. Gheisi AR, Neygandhi C, Sternig AK, Carrasco E, Marbach H, Thomele D, Diwald O. O2 adsorption dependent photoluminescence emission from metal oxide nanoparticles. Phys Chem Chem Phys 2014;16:23922e9. https://doi.org/10.1039/c4cp03080j. 19. Eliseeva SV, Bunzli€ J-CG. Lanthanide luminescence for functional materials and bio- sciences. Chem Soc Rev 2009;39:189e227. https://doi.org/10.1039/B905604C. 20. Chambers MD, Clarke DR. Doped oxides for high-temperature luminescence and lifetime thermometry. Annu Rev Mater Res 2009;39:325e59. https://doi.org/10.1146/ annurev-matsci-112408-125237. 21. Gabriel S, Elhanan W. Energy gap law in the solvent isotope effect on radiationless transitions of rare earth ions. J Chem Phys 1975;62:208e13. 22. Nagai H, Noguchi Y. Ambient gas influence on photoluminescence intensity from InP and GaAs cleaved surfaces. Appl Phys Lett 1978;33:312e4. https://doi.org/10.1063/ 1.90351. 23. Siedl N, Koller D, Sternig AK, Thomele D, Diwald O. Photoluminescence quenching in compressed MgO nanoparticle systems. Phys Chem Chem Phys 2014;16:8339e45. https://doi.org/10.1039/c3cp54582b. 24. Mochizuki S, Araki H. Reversible photoinduced spectral transition in Eu2O3-gAl2O3 composites at room temperature. Phys B Condens Matter 2003;340e342:913e7. 25. Reedo V, Lange S, Kiisk V, Lukner A, Tatte T, Sildos I. Influence of ambient gas on the 3þ photoluminescence of sol-gel derived TiO2:Sm films. In: Rosental A, editor. Optical materials and applications: 4th international conference on advanced optical materials and devices. SPIE; 2005. pp. 59460F 1e6. SPIE 2005 5946 59460F.pdf. 26. Eltermann M, Lange S, Utt K, Joost U, Kink I, Kiisk V, Sildos I. TiO2 nano-powders as potential low-temperature optical gas sensors. Key Eng Mater 2014;605:368e71. https://doi.org/10.4028/www.scientific.net/KEM.605.368. 3þ 27. Eltermann M, Utt K, Lange S, Jaaniso R. Sm doped TiO2 as optical oxygen sensor material. Opt Mater 2015;51:24e30. https://doi.org/10.1016/j.optmat.2015.11.020. 28. Eltermann M, Kiisk V, Berholts A, Dolgov L, Lange S, Utt K, Jaaniso R. Modeling of luminescence-based oxygen sensing by redox-switched energy transfer in nanocrystal- þ line TiO :Sm3 . Sensor Actuator B Chem 2018;265:556e64. 2 þ þ 29. Luo W, Li R, Chen X. Host-sensitized luminescence of Nd3 and Sm3 ions incorporated in anatase titania nanocrystals. J Phys Chem C 2009;113:8772e7. 302 V. Kiisk and Raivo Jaaniso

30. Chakraborty A, Debnath GH, Saha NR, Chattopadhyay D, Waldeck DH, Mukherjee P. Identifying the correct hosteguest combination to sensitize trivalent lanthanide (guest) luminescence: titanium dioxide nanoparticles as a model host system. J Phys Chem C 2016;120:23870e82. https://doi.org/10.1021/ acs.jpcc.6b08421. 31. Almeida NAF, Rodrigues J, Silva P, Emami N, Soares MJ, Monteiro T, Lopes-da- Silva JA, Marques PAAP. Pressure dependent luminescence in titanium dioxide particles modified with europium ions. Sensor Actuator B Chem 2016;234:137e44. https:// doi.org/10.1016/j.snb.2016.04.157. 32. Tachikawa T, Ishigaki T, Li J-G, Fujitsuka M, Majima T. Defect-mediated photolumi- 3þ nescence dynamics of Eu -doped TiO2 nanocrystals revealed at the single-particle or single-aggregate level. Angew Chem Int Ed 2008;47:5348e52. https://doi.org/10.1002/ anie.200800528. 33. qojkowski W, Ga1a˛zka K, Opalinska A, Chudoba T, Swiderska-Sroda A, Millers D, Grigorjeva L, Smits K. Method of measuring of oxygen content in gas. 2012. WO 2012110967 A1 (EP2686669 A1). 34. Puust L, Kiisk V, Eltermann M, M€andar H, Saar R, Lange S, Sildos I, Dolgov L, Matisen L, Jaaniso R. Effect of ambient oxygen on the photoluminescence of sole gel-derived nanocrystalline ZrO2:Eu,Nb. J Phys D Appl Phys 2017;50:215303. https://doi.org/10.1088/1361-6463/aa6c48. 35. Kiisk V, Puust L, M€andar H, Ritslaid P, Bite I, Jankovica D, Sildos I, Jaaniso R. Phase stability and oxygen-sensitive photoluminescence of ZrO2:Eu,Nb nanopowders. Mater Chem Phys 2018;214:135e42. 36. Di W, Wang X, Xinguang R. Nanocrystalline CePO4 :Tb as a novel oxygen sensing material on the basis of its redox responsive reversible luminescence. Nanotechnol 2010;21:075709. þ 37. Tang W, Sun Y, Wang S, Du B, Yin Y, Liu X, Yang B, Cao W, Yu M. Pr3 -Doped (K0.5Na0.5)NbO3 as a high response optical oxygen sensing agent. J Mater Chem C 2016; 4:11508e13. https://doi.org/10.1039/C6TC04216C. 38. Weng H, Xu X-Y, Yan B. Novel multi-component photofunctional nanohybrids for ratio-dependent oxygen sensing. J Colloid Interface Sci 2017;502:8e15. https://doi.org/ 10.1016/j.jcis.2017.04.081. þ 39. Van den Eeckhout K, Smet PF, Poelman D. Persistent luminescence in Eu2 -doped compounds: a review. Materials 2010;3:2536e66. 40. Aydin I, Ertekin K, Demirci S, Gultekin S, Celik E. Sol-gel synthesized Sr Al O : þ þ 4 14 25 Eu2 /Dy3 blueegreen phosphorous as oxygen sensing materials. Opt Mater 2016; 62:285e96. https://doi.org/10.1016/j.optmat.2016.10.019. 41. Shannon RD, Pask JA. Kinetics of the anatase-rutile transformation. J Am Ceram Soc 1965;48:391e8. https://doi.org/10.1111/j.1151-2916.1965.tb14774.x. 42. Reidy DJ, Holmes JD, Morris MA. The critical size mechanism for the anatase to rutile transformation in TiO2 and doped-TiO2. J Eur Ceram Soc 2006;26:1527e34. 43. Setiawati E, Kawano K. Stabilization of anatase phase in the rare earth; Eu and Sm ion doped nanoparticle TiO2. J Alloys & Comp 2008;451:293e6. 44. Tang H, Prasad K, Sanjines R, Schmid PE, Lévy F. Electrical and optical properties of TiO2 anatase thin films. J Appl Phys 1994;75:2042e7. 45. Almquist CB, Biswas P. Role of synthesis method and particle size of nanostructured TiO2 on its photoactivity. J Catal 2002;212:145e56. https://doi.org/10.1006/ jcat.2002.3783. 46. Frindell KL, Bartl MH, Robinson MR, Bazan GC, Popitsch A, Stucky GD. Visible and near-IR luminescence via energy transfer in rare earth doped mesoporous titania thin films with nanocrystalline walls. J Solid State Chem 2003;172:81e8. Rare earthedoped oxide materials for photoluminescence-based gas sensors 303

47. Dolgov L, Eltermann M, Lange S, Kiisk V, Zhou L, Shi J, Wu M, Jaaniso R. Au/SiO2 3þ nanoparticles in TiO2:Sm films for improved fluorescence sensing of oxygen. J Mater Chem C 2017;5:11958e64. https://doi.org/10.1039/C7TC03704J. 48. Kiisk V, Savel M, Reedo V, Lukner A, Sildos I. Anatase-to-rutile phase transition of þ samarium-doped powder detected via the luminescence of Sm3 . Physics Procedia 2009;2:527e38. 49. Tang H, Berger H, Schmid PE, Lévy F. Photoluminescence in TiO2 anatase single crystals. Solid State Commun 1993;87:847e50. þ 50. Lange S, Sildos I, Kiisk V, Aarik J. Energy transfer in the photoexcitation of Sm3 - implanted TiO2 thin films. Mater Sci Eng B 2004;112:87e90. 51. Blumen A, Manz J. On the concentration and time dependence of the energy transfer to randomly distributed acceptors. J Chem Phys 1979;71:4694e702. https://doi.org/ 10.1063/1.438253. 52. Tang H, Lévy F, Berger H, Schmid PE. Urbach tail of anatase TiO2. Phys Rev B 1995; 52:7771e4. 53. Yamakata A, Ishibashi T, Onishi H. Time-resolved infrared absorption spectroscopy of photogenerated electrons in platinized TiO2 particles. Chem Phys Lett 2001;333:271e7. https://doi.org/10.1016/S0009-2614(00)01374-9. 54. Antila LJ, Santomauro FG, Hammarstrom€ L, Fernandes DLA, Sa J. Hunting for the elusive shallow traps in TiO2 anatase. Chem. Commun. 2015;51:10914e6. https:// doi.org/10.1039/C5CC02876K. 55. Setiawati E, Kawano K, Tsuboi T, Seo HJ. Studies on thermal migration of Eu ion doped into TiO2 nanoparticles. Jpn J Appl Phys 2008;47:4651. 56. Intrinsic defects of TiO2(110): interaction with chemisorbed O2,H2, CO, and CO2. Phys Rev B 1983;28:3427. 57. Linsebigler AL, Lu G, Yates JT. Photocatalysis on TiO2 surfaces: principles, mechanisms, and selected results. Chem Rev 1995;95:735e58. 58. Batzill M, Diebold U. Surface studies of gas sensing metal oxides. Phys Chem Chem Phys 2007;9:2307e18. https://doi.org/10.1039/b617710g. 59. Henrich VE. The surfaces of metal oxides. Rep Prog Phys 1985;48:1481. 60. Munuera G, Rives-Arnau V, Saucedo A. Photo-adsorption and photo-desorption of oxygen on highly hydroxylated TiO2 surfaces. Part 1.drole of hydroxyl groups in photo-adsorption. J Chem Soc, Faraday Trans 1979;1(75):736e47. https://doi.org/ 10.1039/F19797500736. 61. Yates Jr JT. Photochemistry on TiO2: mechanisms behind the surface chemistry. Surf Sci 2009;603:1605e12. 62. Nelson J, Eppler AM, Ballard IM. Photoconductivity and charge trapping in porous nanocrystalline titanium dioxide. J Photochem Photobiol A Chem 2002;148:25e31. 63. Graeve OA. Zirconia. In: Ceramic and glass materials: structure, properties and processing. Springer; 2008. p. 169e97. 64. Fabris S, Paxton AT, Finnis MW. A stabilization mechanism of zirconia based on oxygen vacancies only. Acta Mater 2002;50:5171e8. https://doi.org/10.1016/S1359- 6454(02)00385-3. 65. Garvie RC. The occurrence of metastable tetragonal zirconia as a crystallite size effect. J Phys Chem 1965;69:1238e43. 66. Djurado E, Bouvier P, Lucazeau G. Crystallite size effect on the tetragonal-monoclinic transition of undoped nanocrystalline zirconia studied by XRD and Raman spectrometry. J Solid State Chem 2000;149:399e407. https://doi.org/10.1006/ jssc.1999.8565. 67. Balog M, Schieber M, Michman M, Patai S. Chemical vapor deposition and characterization of HfO2 films from organo-hafnium compounds. Thin Solid Films 1977;41: 247e59. 304 V. Kiisk and Raivo Jaaniso

68. Kiisk V, Puust L, Utt K, Maaroos A, M€andar H, Viviani E, Piccinelli F, Saar R, Joost U, Sildos I. Photo-, thermo- and optically stimulated luminescence of monoclinic zirconia. J Lumin 2016;174:49e55. https://doi.org/10.1016/j.jlumin.2015.12.020. 69. Fidelus JD, Lojkowski W, Millers D, Smits K, Grigorjeva L. Advanced nanocrystalline ZrO2 for optical oxygen sensors. In: 8th IEEE conference on sensors; 2009. p. 1268e72. 70. Mochizuki S, Fujishiro F. The photoluminescence properties and reversible photoinduced spectral change of CeO2 bulk, film and nanocrystals. Phys. Stat. Sol. (b) 2009; 246:2320e8. https://doi.org/10.1002/pssb.200844419. 71. Mochizuki S, Saito T. Defect-effects on the photoluminescence of ZrO2 bulk, film and nanocrystals. Phys B Condens Matter 2012;407:2911e4. 72. Smits K, Sarakovskis A, Grigorjeva L, Millers D, Grabis J. The role of Nb in intensity increase of Er ion upconversion luminescence in zirconia. J Appl Phys 2014;115: 213520. https://doi.org/10.1063/1.4882262. 73. Smits K, Olsteins D, Zolotarjovs A, Laganovska K, Millers D, Ignatans R, Grabis J. Doped zirconia phase and luminescence dependence on the nature of charge compensation. Sci Rep 2017;7:44453. https://doi.org/10.1038/srep44453. 74. Shannon RD. Revised effective ionic radii and systematic studies of interatomic distances in halides and chaleogenides. Acta Crystallogr n.d.;32:751e67. 75. Smits K, Jankovica D, Sarakovskis A, Millers D. Up-conversion luminescence dependence on structure in zirconia nanocrystals. Opt Mater 2013;35:462e6. 76. Li L, Yang HK, Moon BK, Choi BC, Jeong JH, Jang K-W, Lee HS, Yi SS. Structure, charge transfer bands and photoluminescence of nanocrystals tetragonal and monoclinic ZrO2:Eu. J Nanosci Nanotechnol 2011;11:350e7. 77. Lange S, Kiisk V, Aarik J, Kirm M, Sildos I. Luminescence of ZrO2 and HfO2 thin films implanted with Eu and Er ions. Phys Status Solidi 2007;4:938e41. 78. Synthesis and characterization of mixed-morphology CePO4 nanoparticles. J Solid State Chem 2007;180:840e6. https://doi.org/10.1016/j.jssc.2006.12.009. 79. Fang Y-P, Xu A-W, Song R-Q, Zhang H-X, You L-P, Yu JC, Liu H-Q. Systematic synthesis and characterization of single-crystal lanthanide orthophosphate nanowires. J Am Chem Soc 2003;125:16025e34. https://doi.org/10.1021/ja037280d. 80. Hikichi Y, Nomura T, Tanimura Y, Suzuki S, Miyamoto M. Sintering and properties of monazite-type CePO4. J Am Ceram Soc 1990;73:3594e6. https://doi.org/10.1111/ j.1151-2916.1990.tb04263.x. 81. Kamiya S, Mizuno H. Phosphors for lamps. In: Yen WM, Shionoya S, Yamamoto H, editors. Phosphor handbook. 2nd ed. CRC Press; 2007. 3þ 3þ 82. Zhu H, Zhu E, Yang H, Wang L, Jin D, Yao K. High-brightness LaPO4:Ce ,Tb nanophosphors: reductive hydrothermal synthesis and photoluminescent properties. J Am Ceram Soc 2008;91:1682e5. https://doi.org/10.1111/j.1551-2916.2008.02320.x. 83. Buissette V, Moreau M, Gacoin T, Boilot J-P. Luminescent core/shell nanoparticles 3þ with a rhabdophane LnPO4-xH2O structure: stabilization of Ce -doped compositions. Adv Funct Mater 2006;16:351e5. https://doi.org/10.1002/ adfm.200500285. 84. Riwotzki K, Meyssamy H, Kornowski A, Haase M. Liquid-phase synthesis of doped nanoparticles: Colloids of luminescing LaPO4:Eu and CePO4:Tb particles with a narrow particle size distribution. J Phys Chem B 2000;104:2824e8. https://doi.org/ 10.1021/jp993581r. 85. Tang W, Wang S, Li Z, Sun Y, Zheng L, Zhang R, Yang B, Cao W, Yu M. Ultrahigh- 3þ sensitive optical temperature sensing based on ferroelectric Pr -doped (K0.5Na0.5) NbO3. Appl Phys Lett 2016;108:061902. https://doi.org/10.1063/1.4941669. 86. Wu J, Xiao D, Zhu J. Potassiumesodium niobate lead-free piezoelectric materials: past, present, and future of phase boundaries. Chem Rev 2015;115:2559e95. https:// doi.org/10.1021/cr5006809. Rare earthedoped oxide materials for photoluminescence-based gas sensors 305

þ 87. Rani J, Patel PK, Adhlakha N, Singh H, Yadav KL, Prakash S. Mo6 modiﬁed (K0.5Na0.5)NbO3 lead free ceramics: structural, electrical and optical properties. J Mater Sci Technol 2014;30:459e65. https://doi.org/10.1016/j.jmst.2013.10.022. 88. Jiang H, Su TT, Gong H, Zhai YC. Direct preparation of K0.5Na0.5NbO3 powders. Cryst Res Technol 2011;46:85e9. https://doi.org/10.1002/crat.201000501. 89. Boutinaud P, Mahiou R, Cavalli E, Bettinelli M. Red luminescence induced by inter- þ valence charge transfer in Pr3 -doped compounds. J Lumin 2007;122e123:430e3. https://doi.org/10.1016/j.jlumin.2006.01.198. This page intentionally left blank PART THREE

Methods and integration

307j This page intentionally left blank CHAPTER TEN

Recent progress in silicon carbide ﬁeld effect gas sensors

M. Andersson, A. Lloyd Spetz, D. Puglisi Linkoping€ University, Linkoping,€ Sweden

Contents

10.1 Introduction 309 10.2 Background: transduction and sensing mechanisms 312 10.2.1 Transducer platform 313 10.2.2 Transduction mechanisms 316 10.2.3 Sensing mechanisms 318 10.2.3.1 General 318 10.2.3.2 Detection of hydrogen-containing gases 320 10.2.3.3 Detection of nonhydrogen-containing gases 324 10.3 Sensing layer development for improved selectivity of SiC gas sensors 327 10.3.1 New material combinations 327 10.3.2 Tailor-made sensing layers for oxygen 328

10.3.3 Tailoring layers for CO2 and NOx 329 10.4 Dynamic sensor operation and advanced data evaluation 332 10.5 Applications 335 10.5.1 Sensor packaging 335 10.5.2 Applications and ﬁeld tests 336 10.6 Summary 338 Acknowledgments 339 References 339

10.1 Introduction Chemical sensors based on silicon field effect transistors (Si-FETs) were introduced in the 1970s when, first, the ion-sensitive FET for pH measure- ments1 and, in 1974, the hydrogen-sensitive metal oxide semiconductor (MOS) FET2,3 were invented. After more than 40 years of research and development on chemical gas sensors, today the field effect transistor gas sensor based on silicon carbide (SiC-FET) is recognized as the most suitable for detection of a variety of different gas molecules at operating temperatures

e from about 200 Ctomorethan600C.4 8 Regardless of Si or SiC as the semiconductor in the field effect gas sensor, a catalytically active gate material such as palladium (Pd), platinum (Pt), or iridium (Ir) provides its gas sensitivity. Besides transistors, MOS capacitors and Schottky diodes with catalytic gate contacts have been developed for gas-sensing purposes, the basic sensing mechanism being common to all the different field effect sensor devices. On exposure of the sensors to a certain substance or gas mixture, the interaction between the gas and gate contact changes the electrical field across the MOS structure, in turn modulating the current through, or the capacitance over, the device. The introduction of the first metal insulator semiconductor (MIS) gas sensor devices based on SiC in the early 1990s9,10 opened up for new applications of field effect sensors. In 1999, at the International Conference of Silicon Carbide and Related Materials in North Carolina, USA, the first gas sensor based on a SiC-FET was presented,11 and one of the main results of this development is devices with excellent long-term stability.7 The wide bandgap of SiC (3.26 eV for the commonly used polytype 4H) permits operational temperatures beyond the limit of approximately 200 C for Si-based sensors without suffering intrinsic conduction effects. Extending the range of sensor operation temperatures allowed exploration of gas metal interactions and catalytic reactions occurring above 200 C, facilitating detection of many more compounds. SiC is also chemically inert, preventing device degradation caused by high temperature or reactions with other materials or substances. SiC-based field effect sensors have therefore been utilized in high temperature (up to 600 C) and corrosive applications such as combustion control in car exhausts and small- and medium-scale 12e15 power plants, monitoring of ammonia (NH3) slip from diesel exhaust and flue gas after treatment systems,16,17, as well as for indoor air quality e control.18 20 Commercial sensor systems based on SiC are available through an SME launched in 2007 (SenSiC AB, Kista, Stockholm, Sweden, www.sensic.se). Monitoring the regeneration of nitrogen oxides (NOx) storage catalysts has also been suggested as an application suitable for field effect sensors based on SiC.21 Olga Casals et al. investigated SiC-based MIS capacitors with Pt/TaOx gate metal in an atmosphere with high relative humidity, 45% RH. Detection of 1 part per million (ppm) hydrogen (H2) at 260 C, 2 ppm carbon monoxide (CO) at 240 C, and 20 ppm ethene (C2H4) at 320 C in nitrogen (N2) was possible even at this high humidity level, and variation in the humidity (15%e45%) did not influence the response. The authors conclude that the SiC sensors Recent progress in silicon carbide field effect gas sensors 311 are especially suitable for monitoring exhaust gases from hydrogen or hydrocarbon-based fuel cells.22 Field effect devices based on other wide bandgap semiconductorsdsuch as diamond,23 gallium nitride (GaN), and aluminum gallium nitride (AlGaN)24,25dhave also been demonstrated for gas-sensing purposes. Chen et al. fabricated a Schottky diode based on GaN on sapphire substrate with a Pd nanoparticleemodified top layer on the Pd gate contact. The detection limit for H2 in air is less than 0.8 ppm at 25 C; however, there is an influence of the humidity level at this low temperature.26 Chou et al. fabricated two Pd/AlGaN/GaN Schottky diodes on sapphire, one with pyramid-like Pd nanoparticles on top and one without.27 The pyramid nanoparticles improved the sensitivity to H2 with a detection limit of 10 parts per billion (ppb) in air at 27 C. Guo et al. reported ultralow electrostatic detection of trinitrotoluene from 0.1 parts per trillion (ppt) to 10 ppb in buffer solution using an AlGaN/GaN high electron mobility transistor with gold nanoparticles functionalized with cysteamine.28 Offermans et al. demonstrated AlGaN/GaN two-dimensional electron gas (2DEG) devices, which are processed as suspended membranes on Si29 and operated with ultralow power. Without additional sensing material, detection of nitrogen dioxide (NO2) concentrations between 11 and 20 ppb in single ppb steps is demonstrated at 250 C with low influence of humidity, and at 275 Cammonia (1e12 ppm in humid atmosphere) is detected in the opposite direction. By adding Pt as the gate contact, H2 is detected at 150 C for a concentration range of 300e3000 ppm. When applying a sensing layer of a pH-sensitive polymer that retains water, CO2 formed charged species in the liquid phase and could be detected at 25 C in the range 1000e6000 ppm. Weng et al. fabricated MISiC capacitors with a gate contact of Pd/TiO2 on top of oxidized SiC (Pd/TiO2/SiO2/SiC). At 325 C, in a mixture of H2 and oxygen (O2), the response to H2 is lower as compared with the response to H2 in N2. On the other hand, a mixture of hydrogen sulfide (H2S) and O2 30 gives a larger response as compared with H2SinN2. This is due to the Claus reaction,31 according to which the oxygen in the presence of titanium dioxide (TiO2) reacts with the sulfur in the H2S molecule and accordingly both hydrogen atoms are released and may participate in the detection process. Nakagomi et al. compared Schottky diodes based on the polytype 4H-SiC with a low doped epilayer and b-gallium(III) oxide (b-Ga2O3) with Pt-sensing electrodes and Ni/Pt or Ti/Al/Pt/Au as the ohmic contacts on the rear side. The Pt-Ga2O3 showed a lower detection limit for hydrogen in oxygen, at 400 C of a few tens ppm of hydrogen. Nonstoichiometry conditions of 312 M. Andersson et al.

the Ga2O3 surface or of the Pt-Ga2O3 interface is suggested as the reason for this.32 In their next paper,33 two devices in series are processed on 1 mm b-Ga2O3 on sapphire, one with Pt gate electrode, the other device without gate electrode and with ohmic contacts as above. It is demonstrated that this design allows stable hydrogen detection in oxygen atmosphere from about 40 ppm even with temperature fluctuations as large as 150 C in the temperature range 400e550 C. For lower temperatures, the resistivity of the Ga2O3 is too large and for higher temperatures the Ga2O3 itself is sensitive to hydrogen. The crystal structure b-Ga2O3 is an n-type material, with a band gap equal to 4.9 eV. Thin films of b-Ga2O3 were deposited on top of p-type nickel(II) oxide (NiO) substrates whereby an interfacial layer 34 of g-Ga2O3 was found between the two materials. The last decade has seen the development of new devices, new material combinations, and new operation modes of SiC-based field effect sensors, and, with the advent of epitaxially grown graphene on SiC, as well as with the integration of a number of 2D materials for gas sensing applications,35 the field is expanding even further. Very promising possibilities for ultrasensitive detection of gaseous compounds are offered by epitaxially grown graphene e on SiC-based sensor structures,36 39 as reported in the first edition of this book.40 Nanoparticle decoration of the graphene surface has considerably improved selectivity, sensitivity, and speed of response of graphene sensors, e while the intrinsic properties of graphene were retained.41 43 This area now also expands into development of novel 2D materials on SiC other than graphene for gas- and liquid-phase sensing applications.44,45 Based on theoretical modeling and material research, selectivity and sensitivity toward various gases are currently also being improved by the development of new combinations of gas-sensitive layers. Recent trends, reviewed in this chapter, include simplification of device designs to reduce fabrication costs and increase stability, as well as novel designs to facilitate sensor packaging, high reusability, and thus an efficient product development. Especially high temperature applications require advanced packaging solutions and a novel approach using low temperature cofired ceramic (LTCC) is presented. Dynamic sensor operation through temperature and gate bias cycling is another recent line of development that makes use of advanced data evaluation to enhance stability, as well as selectivity toward certain substances.

10.2 Background: transduction and sensing mechanisms In this section, the basic physical principles and electrical operation of the transducer platform, the FET device, are given. Moreover, a description Recent progress in silicon carbide field effect gas sensors 313 of the sensing mechanisms, when the devices are used as gas sensors, is given in general and for hydrogen- and nonhydrogen-containing gases. For this section, we also refer to Ref. 7. 10.2.1 Transducer platform The MIS capacitor represents the heart of most field effect sensor devices, and the physics of MIS capacitors has been widely studied and treated in e detail in well-known semiconductor physics and other sensor books.46 48 Here, we will only give the basic physical principles regarding the metal insulator semiconductor field effect transistor (MISFET), because this is the ultimate transducer for commercial sensor devices. MISFET devices may be distinguished in normally off or enhancement type and normally on or depletion type devices. Normally off means that with zero applied gate bias no channel between drain and source is created, whereas normally on means that a channel already exists at zero applied gate bias. More details can be found in Ref. 49. A schematic of the enhancement type MISFET device under different conditions and its corresponding currentevoltage (I/V) characteristics is shown in Fig. 10.1. The channel conductance, determined by its dimensions, the mobility of the electrons, and the inversion charge density of electrons can be modulated by the gate bias, VGS. When no gate bias is applied (VGS ¼ 0), there is no conductive path from source to drain, therefore no current flows through the conducting channel (Fig. 10.1(a)). As soon as a gate bias is applied (VGS > 0), the channel (n-type inversion layer) develops allowing electrons to flow between the source and drain terminals in response to a drain bias (VDS). For a gate bias larger than the threshold voltage VT (VGS > VT) and small VDS, the device operates in the so-called linear region (Fig. 10.1(b)). As the drain-source voltage increases, the voltage drop across the insulator near the drain terminal decreases. This means that the induced inversion charge density near the drain decreases, the channel depth (i.e., the thickness of the inversion channel) near the drain terminal is reduced, and the slope of the I/V curve decreases. The point at which the channel depth at the drain is reduced to zero is called pinch-off and represents the onset of saturation (Fig. 10.1(c)). Here, the voltage drop across the insulator at the drain is equal to the threshold voltage (VDS,sat ¼ VGSeVT). Beyond the pinch-off point, the drain-source current remains constant, resulting in a flat I/V curve (Fig. 10.1(d)). This region is called saturation region. 314 M. Andersson et al.

(a) +VGS +VDS V < V ID GS T G S Insulator D n+ n+ p-type epilayer n-type substrate

VDS

(b) +VGS V < V small V +VDS ID GS T, DS G S Insulator D VGS2 > VGS1 n+ n+ VGS2 p-type epilayer V n-type substrate GS1

VDS

+V (c) GS +V DS ID VGS > VT, VDS = VD,sat G S Insulator D VDS,sat (VGS2) n+ n+ p-type epilayer V (V ) n-type substrate DS,sat GS1

VDS

(d) +VGS +V I DS D VGS > VT, VDS > VDS,sat G S Insulator D n+ n+ p-type epilayer n-type substrate

VDS Figure 10.1 Enhancement type metal insulator semiconductor ﬁeld effect transistor (MISFET) device under different operating conditions and corresponding I/V curves.

MISFET operated (a) in equilibrium condition (VGS ¼ 0), (b) in the linear region, (c) at the onset of saturation, and (d) beyond saturation.

Transistor-based sensor devices are commonly operated in saturation mode. The drain current (ID,sat) versus gate voltage (VGS) relationship for the saturation region is described quantitatively by m ε W n ins 2 ID;sat ¼ ½VGS VT (10.1) 2Ldins Recent progress in silicon carbide ﬁeld effect gas sensors 315

1=2 2dins½eNaεsFF Qssdins VT ¼ þ Fms þ 2FF (10.2) εins εins where W and L are the channel width and length, respectively; mn is the channel electron mobility; εins and dins are the insulator permittivity and thickness, respectively; VT is the threshold voltage; e is the elementary charge; Na the bulk doping concentration; εs the semiconductor permittivity; QSS the insulator charge density; Fms the metal-to-semiconductor work function difference; and FF is the Fermi potential, which is the potential difference between the Fermi level and the intrinsic Fermi level. Regarding the transistor-based sensor devices, enhancement and depletion type MISFET transistors are both used. A detailed study of the difference between enhancement and depletion type SiC-FET gas sensors can be found in Ref. 7. However, the depletion type MISFET has the advantages of operation at zero or very low applied gate voltage, less influence of temperature fluctuations, and generally more stable operation of the sensors, therefore it is preferable as a gas sensor. Concerning the depletion type MISFET, even when no bias is applied to the gate terminal (VGS ¼ 0), there is a current flow, i.e., a conductive path from source to drain exists. The threshold voltage of this device is defined by the difference between the built-in voltage across the gate metal/insulator/SiC stack and the pinch-off voltage. The latter is the applied voltage which cuts off the conducting path between source and drain and is dependent on the thickness, ds, and the doping level, Nd,of the n-type active layer:

eNdds Vp ¼ (10.3) 2εs Eq. (10.3) shows the pinch off voltage of the depletion type MISFET, εs is the permittivity of the semiconductor. For a more in-depth treatment of field effect device theory and operation, see, for instance Refs. 7, 46e48. The design of the device parameters influences the size of the gas response. It has been demonstrated, in the case of a SiC-FET with porous Ir as the gate material, that a decrease of the gate length from 40 to 20 mm results in a factor two increase of the sensor response to CO in 3% oxygen at an operating temperature of 200 C.7 Other studies showed that optimizing the thickness of the gate dielectric almost doubled the gas response to ammonia. In addition, optimizing the device for lower field 316 M. Andersson et al. strength between the different terminals of the device increased the long- term performance. 10.2.2 Transduction mechanisms Parameters such as device dimensions, electron mobility, permittivity, and doping concentration are inherent to the choice of materials, the design, and the processing of field effect sensor devices. Once fabricated, the values of these are fixed but the charges located in or at the surface of the insulator, QSS, the metal-to-semiconductor work function difference, Fms, and any internal gate voltage drop, VGSint, added to the externally applied gate bias, VGSext, can also have an influence on the drain current, ID. Any change in the values of one or more of these parameters will change the I/V characteristics of the FET devices. Thus, if the interactions between the gas and gate materials on exposure to a certain substance lead to the introduction of an internal gate voltage drop, a change in gate insulator charge, and/or a change in gate metal work function, the substance could be detected through a change in drain current (see Fig. 10.2). This requires the injection of charge to or charge separation at the gate contact/insulator interface, or species capable of changing the metal work function to adsorb on the inner surface of the gate contact material. Examples of changes in ID/VGS characteristics for a gas-induced internal voltage drop are given in Fig. 10.2(c). When atoms or molecules adsorb on a surface, there is most often some kind of charge transfer between the adsorbates and the surface, and thus a separation of charge, as well as a change in work function of the material. One kind of field effectebased gas sensor, the suspended gate FET (SGFET), utilizes this latter phenomenon.50 The design of SGFETs includes a very small air gap between the gate contact material and the insulator, just large enough to facilitate rapid diffusion of gas molecules to the gate contact surface facing the insulator. Any change in drain current, i.e., sensor signal, on gas exposure is directly related to the change in work function of the gate material, resulting from adsorption of one or more gaseous substances to its surface. As mentioned, a common mode of operation of transistor-based field effect sensors is to keep the drain current constant and measure the resulting drain-source voltage drop as a sensor signal. Connecting the transistor’s drain and gate (enhancement-type devices) or source and gate (depletion-type devices) terminals, when operating the device as a gas sensor, makes it a simple two-terminal device (e.g. Ref. 7). In the other Recent progress in silicon carbide field effect gas sensors 317

(a) NH3 CO H2O CO2 H2 CO O2 NO NH3 NO2 H2 O2 NO – – – O – –O __ _ _ _ O O O O– O– HH+ + H + HH + + Insulator Insulator

p-type epilayer p-type epilayer

(b) eVins eVins Ec Ec

EFi Gas EFi _ eΦs + EF eΦs EF Exposure _ eVGS,ext + Ev Ev eVGS,int

MetalM InsulatorI SemiconductorS MetalM InsulatorI SemiconductorS (c)

ID VGS,ext = VDS ID VGS,ext + VGS,int = VDS

VGS = 4V VGS = 5V VGS = 3V VGS = 4V

VGS = 3V VGS = 2V VGS= 1V VGS = 2V

1 2 3 4 5 VDS [V] 1 2 3 4 5 VDS[V] Figure 10.2 (a) Examples of reactions on the catalytic metal gates are displayed, as well as the effect of hydrogen and oxygen anion adsorption on the number of charge carriers in the channel. (b) Corresponding changes of the energy band diagram, air/ inert atmosphere to the left and hydrogen exposure to the right and (c) the change in I/V characteristics following hydrogen exposure. mode of operation, the drain current is measured as the sensor signal at a constant drain-source voltage. The choice of the operation mode, at a constant drain current or at a constant drain-source voltage, as well as of the electrical operating point along the currentevoltage (I/V) curve of the device also inﬂuences the size of the gas response. As an example, we demonstrated that operating a SiC-FET sensor, with porous Ir on top of a dense thin ﬁlm of tungsten trioxide (WO3) as the sensing layer, at a constant drain- source voltage and measuring the drain current as the sensor signal, gave a sensor response to 100 ppb benzene which was in the saturation region about twice that in the linear region, see Fig. 10.3(a).8 In Fig. 10.3(b),thesame operation mode is used for a SiC-FET with a porous Ir gate, and the detection 318 M. Andersson et al.

Figure 10.3 (a) Sensor response to 10, 50, and 100 ppb of benzene (C6H6) at 300 C, in dry air, and under operation at the linear (upper signal) and saturation (bottom signal) regions of the transistor. (b) Detection limit as a function of relative humidity for formaldehyde (CH2O), benzene (C6H6), and naphthalene (C10H8). For C10H8, the detection limit can only be stated to be below 0.5 ppb, because our gas mixing system cannot provide C10H8 concentrations below 0.5 ppb. limits for formaldehyde, benzene, and naphthalene are shown as a function of relative humidity. For naphthalene, the performance of the gas mixing system sets the limit to 0.5 ppb. 10.2.3 Sensing mechanisms 10.2.3.1 General Work function changes and the creation of internal voltage drops are merely the general mechanisms behind the conversion of chemical interactions between the gas and the sensor device into an electrical output. Voltage drops Recent progress in silicon carbide field effect gas sensors 319 can be introduced and work function changes can be achieved in a number of different ways. To be useful for specific applications, the sensors must, however, be able to distinguish between different gas mixtures and/or quan- tify one or more substances with good resolution. The sensitivity and selectivity toward the substance(s) of interest are important figures of merit for a specific sensor, as are detection limit, speed of response, and stability. The sensor’s sensitivity and selectivity to the analyte of interest are largely determined by the specific interactions between the various ambient gaseous substances and the gate materials exposed to the surrounding gas. These interactions include adsorption and reactions of atoms and molecules on the surfaces of the gate materials, as well as desorption from the same surfaces. In general, adsorption and desorption are dependent on, for example, the ambient temperature, the partial pressure of the substance, the desorption energy, and the sticking coefficient. The sticking coefficient gives the probability for adsorption of a molecule incident on an empty adsorption site and is dependent on temperature and activation energy for adsorption. It is therefore different for different molecules, surface compositions, and crystal orientations. Furthermore, the adsorption of molecules on the sensor surface may be direct or, via precursor states, it may be dissociative or nondissociative and there may be interactions between adsorbed species on the surface. All these details of the adsorption will affect the equilibrium state of the molecules on the sensor surface. In addition, other constituents of the surrounding gas matrix may adsorb to the surface and affect the coverage of the target substance in different ways (e.g., by reducing or blocking adsorption of this substance or removing it from the surface through chemical reactions). At the steady state, equilibrium usually develops between the adsorption, chemical surface reactions, and desorption of different substances in the surrounding gas matrix. An overview of the surface processes and examples of their influence on the device characteristics is given in Fig. 10.2.Considering the operational temperature of the sensor to be constant, and the sticking coefficients, interaction, and desorption energies to be inherent to the molecules and the surface, the steady-state condition on the surface is dependent on the partial pressures of the gas matrix constituents and, therefore, reflects the composition of the surrounding gas. Several different gas matrices may, however, give rise to the same equilibrium surface conditions for a certain surface and operational temperature. Conversely, a different surface, or a change in operational temperature, may give rise to a different equilibrium surface condition for the same gas matrix, highlighting the importance and possibilities 320 M. Andersson et al. regarding the choice of gate material and sensor operational temperature, as also exemplified below.

10.2.3.2 Detection of hydrogen-containing gases Hydrogen, H2, adsorbs dissociatively on catalytic gate metals such as Pd, Pt, and Ir. In the presence of hydrogen alone, the steady-state surface coverage of hydrogen atoms follows the simple Langmuir relation51,52 and is only dependent on ambient hydrogen pressure. Normally, also other substances are present in the surrounding atmosphere and may affect the equilibrium coverage of hydrogen in different ways. Notably, oxygen also adsorbs dissociatively on commonly used gate metals at sensor operating temperatures (200e600 C)ethe recombination and desorption rates, however, being very low below 300 C. At this temperature, oxygen can basically be removed at any appreciable rate only through reaction with other atoms or molecules, such as chemisorbed hydrogen in the formation of water. In normal air, the pressure-dependent hydrogen coverage for every gate material is determined by the adsorption and desorption characteristics of hydrogen and its reaction with adsorbed oxygen. Variations in hydrogen and oxygen partial pressures thus lead to a change in hydrogen coverage. The generated hydrogen atoms are, to some extent, also withdrawn from the surface by rapid diffusion through the metal contact to the metal/insulator interface. Due to the very rapid diffusion of the hydrogen atoms, the surface coverage and interface hydrogen concentration are in equilibrium. As concluded from infrared spectroscopy,53 the hydrogen atoms adsorb to oxygen atoms in the surface of oxidic insulators, forming hydroxyl groups (OH) on the oxide accompanied by substantial charge transfer. Because OH groups have a large dipole moment, the interface layer of dipoles introduces a sharp potential step at the interface, earlier referred to as an internal voltage drop, Vint. This voltage drop adds to the externally applied bias, resulting in a shift of the I/V characteristics of the sensor, DVint, as illustrated in Fig. 10.2(c), and is given by: r DVint ¼ nH $ (10.4) ε0 where r is the dipole moment of an OH group, ε0 is the permittivity of free space, and nH is the number of hydrogen atoms per unit area at the interface, which is related to the coverage of hydrogen on the metal surface.54,55 The size of the I/V shift is thus a measure of the ambient partial pressure of hydrogen, in relation to other gases such as oxygen. A corresponding energy Recent progress in silicon carbide field effect gas sensors 321 band diagram illustrating the effect of this dipole layer can be found in Fig. 10.2(b). From a sensor response point of view, it has been shown that dipole formation is the dominant effect regarding hydrogen detection. Work function changes due to adsorption on the metal side of the metal/insulator interface only have a minor influence on the sensor signal, introducing a small shift in the I/V or capacitanceevoltage (C/V) characteristics in the opposite direction to that generated by dipole formation.56 Further evidence for the importance of an oxidic insulator surface has been obtained from sensors based on both SiC and GaN Schottky diodes,57,58 for which the hydrogen response considerably improved on the introduction of a thin oxide between the metal and the semiconductor. When comparing the hydrogen response from devices with different insulator materials (e.g., Al2O3,Ta2O5,SiO2), the response correlates well with the insulator surface density of oxygen atoms,59 further emphasizing the role of the oxygen as adsorption sites for the hydrogen (see Fig. 10.4(a and b)). The choice of insulator thus influences the hydrogen sensitivity of field effect devices, as well as their dynamic range. This was also studied by Roy et al. for capacitive SiC sensors employing either hafnium(IV) oxide (HfO2)orTiO2 as the dielectric and Ti/Pd as the catalytic contact. By multiple linear regression, real-time gas concentration in a mixture of different gas species could be monitored using different catalytic gate metals and different insulators in a sensor array. This work points out that also defects in the insulator surface play a role for gas detection.60 Ofrim et al. also used SiC capacitors as hydrogen sensors utilizing silicon dioxide (SiO2), TiO2, and zinc oxide (ZnO) as the gate insulator, whereby the TiO2-based capacitive sensor showed superior performance.61 In the case of other molecules containing hydrogen, the same basic principles as for hydrogen apply if free hydrogen atoms can be generated on adsorption. At temperatures of approximately 600 Corabove,field effect sensors with catalytic metal gates, here Pt, exhibit a binary response to hydrocarbons, irrespective of hydrocarbon identity (see Fig. 10.4(c)).62 As long as the oxygen concentration is such that complete oxidation of the hydrocarbons can take place on the gate metal, the high reaction rates keep the surface fairly clean from hydrocarbons and, to a large extent, oxygen covered. Any hydrocarbons sticking to the surface are oxidized directly on adsorption without generation of any free hydrogen atoms. When increasing the hydrocarbon concentration 322 M. Andersson et al.

(a) (b)

Al O Al O Si N SiO Al O inert Si N inert SiO inert SiO Ta O inert Response (V) Response (V)

Si N

Hydrogen concentration (ppm) Time (min) (c) (d) 0 600 T=600 ± 4°C –100 –200 –300 –400 500 –500 25 ppm –600 –700 3% O 50 ppm 400 –800 75 ppm butane + 0.5% propane/4%O2 100 ppm 300 butane + 0.5% propane/4%O2 0 125 ppm

V (mV) butane + 1% propane/6%O2 –100 150 ppm Δ butane/1%O2 –200 175 ppm 200 butane/2%O2 –300 200 ppm

Sensor response (mV) –400 butane/4%O2 –500 225 ppm ethane/1%O2 100 –600 10% O 250 ppm ethane/2%O2 –700 ethane/4%O2 –800 0 100 150 200 250 300 350 400 0 1 2 3 4 5 6 α Temperature (°C) Figure 10.4 The figure displays, in (a) and (b), the response to hydrogen in the range 10 ppm to 1% in air or N2 (indicated as inert in (b)) at 140 C for Pd/Pt gate field effect sensors with various insulator materials. In (c) and (d), the response DV to saturated hydrocarbons at 600 C as a function of equivalence ratio a are given, as well as the response to unsaturated hydrocarbons at temperatures of 100e400 C and concentrations well below the equivalence ratio. The equivalence ratio is defined as the ratio of the actual fuel-to-oxidizer ratio to the stoichiometric fuel-to-oxidizer ratio. (a) and (b) are reprinted with permission from the Eriksson M, Salomonsson A, Lundstrom€ I, Briand D, Åbom AE. The influence of the insulator surface properties on the hydrogen response of field-effect gas sensors. J Appl Phy 2005;98(3):34903e8 © 2012 American Institute of Physics. (c) is reprinted with permission from Baranzahi A, Lloyd Spetz A, Glavmo M, Carlsson C, Nytomt J, Salomonsson P, Jobson E, Haggendal€ B, Mårtensson P, Lundstrom€ I. Response of metal-oxide-silicon carbide sensors to simulated and real exhaust gases. Sensor Actuator 1997;B43:52e9. © 1997 Elsevier. (d) is reprinted with permission from the Andersson M, Everbrand L, Lloyd Spetz A, Nystrom€ T, Nilsson M, Gauffin C, Svensson H. A MISiCFET based gas sensor system for combustion control in small-scale wood fired boilers. Proc IEEE Sensors 2007:962e5. Atlanta, USA Oct 28e31. © 2007 IEEE. beyond the stoichiometric hydrocarbon to oxygen ratio, the hydrocarbons reduce the gate metal surface and effectively deplete it of oxygen. Dissociation, rather than oxidation, is the dominating process, producing free hydrogen atoms which can reach the interface and induce an internal voltage drop. Recent progress in silicon carbide field effect gas sensors 323

At temperatures below 300 C, certain hydrocarbonsde.g., unsaturated hydrocarbons such as ethene (C2H4) and propene (C3H6)dmay still reduce the catalytic metal surface and produce free hydrogen even in the presence of excess amounts of oxygen.63,64 The underlying reason is the higher sticking probability of these hydrocarbons compared with oxygen and the lower rates of oxidation at lower temperatures. However, for decomposition of saturated hydrocarbons on the catalytic sensor surface in an atmosphere of excess oxygen, all decomposed hydrogen atoms end up as water molecules, which desorb from the sensor surface. Therefore, no free hydrogen atoms are generated and no sensor response is obtained from these substances for conditions of excess oxygen.64,65 Pt gate sensors operated at 200e300 C therefore also exhibit a binary switch in sensor response to unsaturated hydrocarbons, the switch point being dependent on oxygen concentration and temperature (due to the temperature dependence of the sticking coefficients) (see Fig. 10.3(d)). Binary switch behavior was also reported by Kahng et al. using a SiC capacitor with a dense Pt gate for hydrogen sensing in ultrahigh vacuum (UHV).66 The binary behavior is due to the competition between hydrogen oxidation and diffusion to the metal/oxide interface. It was also concluded that oxygen is needed to restore the sensor baseline after exposure to hydrogen. Another hydrogen-containing substance which has attracted a great deal of interest in the field of high-temperature gas sensors is ammonia (NH3). Ammonia has not been observed to dissociate on adsorption on Pt at temperatures below approximately 225 C.67 Furthermore, there is some evidence of oxygen-mediated dissociation occurring on Pd-MOS sensor devices,68 which leads to direct oxidation of adsorbed NH3. The view that no free hydrogen atoms are generated on the Pt surface accords with the observations from sensors with dense, homogeneous Pt gates, for which no NH3 response is obtained. In case of a discontinuous/porous gate metal (see Fig. 10.2), when exposing parts of the oxide to the ambient atmosphere, the field effect devices exhibit similar sensing characteristics as for hydrogen.69,70 The generally accepted view emphasizes the importance of the three phase boundaries between oxide, metal, and the gas phase as the site for ammonia dissociation to create OH groups on the surface of the oxide.71 At the metal/oxide border, hydrogen from an ammonia molecule may be directly transferred to oxygen atoms in the surface of the oxide, possibly as a proton, the charged complex being stabilized by its proximity to the metal. Fourier transform infrared spectroscopic measurements on a model system consisting of a Pt impregnated SiO2 powder revealed the formation 324 M. Andersson et al.

72 of OH groups at temperatures above 225 C on exposure to NH3. The amount of OH groups formed correlated well with the Pt loading (coverage), which has been interpreted as the formed OH groups being located close to the metal/oxide border. Local response measurements performed on capacitive field effect sensor devices by laterally resolved photocurrent measurements provided similar results, relating the generation of OH groups to the metal/oxide border. Furthermore, these investigations also indicated the possibility for diffusion of hydrogen/protons into the metal/oxide interface underneath the metal,70,71 inducing the same kind of internal voltage drop as in the case of hydrogen exposure. Hydrogen detection sites underneath the (Pt) metal at the metal/insulator interface was systematically studied by Åbom et al. by scratch adhesion measurements, transmission electron microscopy, and atomic force microscopy studies of ripped off metal films.73 The size of the semi-inert hydrogen response increased with roughness of the Pt metal surface facing the insulator, which showed a blocking effect of Pt metal in direct contact with the insulator (SiO2). As previously mentioned, oxygen also adsorbs dissociatively on Pt and negatively charged oxygen atoms may spillover to exposed areas of the oxide surface in devices with a discontinuous (porous) gate contact. At the steady state, an equilibrium between oxygen coverage on the Pt surface and concentration of oxygen anions on the oxide surface would then develop. It has been suggested that the response of porous Pt gate sensors to reducing substances such as hydrogen, hydrocarbons, and ammonia may partly originate from the reverse spillover of oxygen anions and their removal on the Pt surface through reactions with adsorbed hydrogen, hydrocarbon, and ammonia molecules.74,75 It should be noted that the removal of negative charges from the oxide surface has the same effect on the I/V or C/V characteristics of field effect sensors as the voltage drop introduced by OH group formation.

10.2.3.3 Detection of nonhydrogen-containing gases Carbon monoxide (CO) is an example of a reducing, nonhydrogen- containing substance for which the interaction with metal (e.g., Pt) gate ﬁeld effect sensors may cause a substantial change in the I/V or C/V characteristics of a device. Without being able to generate any free hydrogen on adsorption, the CO sensitivity has been stipulated to be at least partly caused by the removal of oxygen anions,74 as discussed above, and/or the reduction e of a surface platinum oxide.76 78 The CO response also correlates well with the CO oxidation characteristics on silica-supported Pt.78,79 At the point Recent progress in silicon carbide ﬁeld effect gas sensors 325

where the oxidation rate suddenly drops when increasing the CO/O2 ratio or decreasing the temperature, the sensor signal exhibits a binary switch from a small to a large response (see also Fig. 10.5).79,80 In analogy with the previously discussed case regarding hydrocarbons, the higher sticking probability of CO compared with oxygen at lower temperatures leads to the Pt

(a) (b) 20912064 1839 0 –100 –200 0.12 –300 –400 CO –500 3% O 0.06 –600 2 0 ppm –700 250 ppm 500 ppm 0.00 0 750 ppm 0.12 –100 1000 ppm

–200 1250 ppm Log (1/R) (a.u.)

Sensor response (mV) –300 0.06 –400 –500 10% O –600 2 0.00 –700 100 150 200 250 300 350 400 2400 2000 1600 Wavenumber (cm–1) (c) Temperature (C°C) (d)

2,00E–009 2,00E–009 (a)CO (a) CO H H 1,00E–009 1,00E–009

0,00E+009 0,00E+009 Partial pressure (Torr) Partial pressure (Torr) Addition of 500 ppm H to the CO pulse 1,6 Addition of 500 ppm H to the CO pulse (b) (b) 1,4 1,4 1,2 1,2 1,0 1,0 0,8 0,8 5% O 0,6 0,6 10% O 0,4 Sensor signal (V) 0,4

Sensor signal (V) 125 250 375 500 625 750 875 100011251250 125 250 375 500 625 750 875 1000 1125 1250 CO concentration in each test gas pulse (ppm) CO concentration in each test gas pulse (ppm)

Figure 10.5 In (a), the CO/O2 and temperature-dependent binary switch of the response of Pt gate field effect sensors toward CO is exemplified, whereas (b) displays the disruption of the adsorbed CO layer on the Pt surface on hydrogen 1 exposure. The spectral peaks at 1839, 2091, and 2064 cm (upper panel; no H2 exposure) correspond to CO adsorbed on Pt, whereas the peaks at wave numbers slightly 1 below 2400 cm (lower panel; exposure to 500 ppm H2 in otherwise the same conditions as in the upper panel) represents gaseous CO2. In (c) and (d), the sensor response toward CO in the range of 125e1250 ppm in the absence/presence of hydrogen (500 ppm) is given for two different oxygen concentrations (lower panels), as well as the downstream H2 and CO2 partial pressures (upper panel). (a) is reprinted with permission from the Andersson M, Everbrand L, Lloyd Spetz A, Nystrom€ T, Nilsson M, Gauffin C, Svensson H. A MISiCFET based gas sensor system for combustion control in small-scale wood fired boilers. Proceedings of the IEEE international conference on sensors, Proc IEEE Sensors 2007:962e5. Atlanta, USA Oct 28e31. © 2007 IEEE. (b) is reprinted with permission from the Becker E, Andersson M, Eriksson M, Lloyd Spetz A, Skoglundh M. Study of the sensing mechanism towards carbon monoxide of platinum-based field effect sensors, IEEE Sens J 2011;11(7):1527e34. © 2011 IEEE. (c) and (d) are reprinted with permission from the Darmastuti Z, Pearce R, Lloyd Spetz A, Andersson M. The influence of gate bias and structure on the CO sensing performance of SiC based field effect sensors. Proc IEEE Sensors 2011:133e6. Limerick, Ireland, October 28e31, 2011. © 2011 IEEE.79 326 M. Andersson et al.

surface being practically covered with CO (unless the CO/O2 ratio is too small), almost excluding oxygen adsorption, also at CO concentrations well below the oxygen concentration. With no or very little oxygen on the surface, the CO oxidation rate is very low. At higher temperature or higher oxygen concentration, the poisoning of the sensor by adsorbed CO on the sensor surface recovers as the CO is removed and the Pt surface rapidly reverts to being dominated by adsorbed oxygen. A large response of Pt gate field effect devices to CO therefore correlates with a surface completely covered by CO, whereas a small CO response is encountered whenever the Pt surface is oxygen dominated. However, not only porous Pt gate contacts exhibit these characteristics. Dense films without any exposed oxide areas also show the same binary switch in sensor signal.81 Furthermore, on introduction of hydrogen at a constant concentration, the large CO response of Pt gate sensors can either increase or decrease, depending on the CO/O2 ratio and temperature (see Fig. 10.5(c and d)). It has also been concluded that the presence of hydrogen can break the self-poisoning of the CO oxidation (see Fig. 10.5(b)).78 This indicates that hydrogen may be able to penetrate/adsorb on a Pt surface covered by CO and, if the CO concentration in relation to the oxygen concentration is small, disrupt the CO coverage. If, instead, the CO/O2 ratio is too high in comparison with the hydrogen concentration, or the Pt surface temperature is too low, the surface remains covered by CO and, effectively, depleted of oxygen. Without any oxygen on the surface, there is no risk of hydrogen adsorbing on the Pt surface being oxidized. A much higher proportion of hydrogen atoms can therefore reach the interface. As a consequence, a CO-covered surface will exhibit a very much larger sensitivity detecting even small concentrations of hydrogen, suggesting the CO response partly being mediated through an increased sensitivity to the background concentration of hydrogen which is present in all gas mixtures. Further support for the influence of hydrogen on the CO response is given from UHV studies on Si-based field effect devices.82,83 As exemplified above, the application-specific performance of a sensor is thus influenced by adsorption, reactions between adsorbed species, diffusion of species on the surface, and desorption characteristics of the individual substances which are present in the gas mixture. These characteristics depend on the materials interacting with the substances, the structure of the materials, and the operating temperature; therefore, the selectivity and sensitivity to the gases of interest can be influenced by the choice of gate materials, their structure, and temperature. For the development of sensors for new Recent progress in silicon carbide field effect gas sensors 327 applications, it is therefore important to gain knowledge about gasesolid interactions and sensor mechanisms to be able to tailor devices with good selectivity and sensitivity to the target substance(s).

10.3 Sensing layer development for improved selectivity of SiC gas sensors The ability of hydrogen atoms to diffuse through the commonly used gate materials renders most of the field effect sensors developed so far to exhibit sensitivity to hydrogen. In addition, nitride-based insulators have a tendency to oxidize over time, providing the necessary sites for hydrogen adsorption.59 In developing sensors for specific applications, the issue of cross-sensitivity to hydrogen and substances containing hydrogen therefore has to be considered. For most applications, this cross-sensitivity has been a limitation for the development of field effectebased devices for sensing of substances that do not contain hydrogen, such as oxygen, nitrogen oxides, and sulfur oxides. To widen the areas of application for field effect sensors by increasing selectivity toward other substances than hydrogen or hydrogen-containing gases, a line of development has been the introduction of new material combinations. However, also the nature of the transducer influences the gas response. SiC-FET devices were studied together with quartz crystal microbalance (QMB) sensors which employed the same porphyrin-based sensing layers. While the SiC-FET device responds to the charging of the gate introduced by the interaction of gases with the porphyrin layer, the QMB device responds (changes of the operating frequency) to the change in total mass of the device due to gas molecules absorbing in the sensing layer.84 Therefore, the combination of the SiC-FET device with the QMB device gives more information about a certain gas mixture. In Section 10.4, we will introduce temperature cycling operation mode and advanced data evaluation to improve selectivity and sensitivity of one sensor working as a virtual sensor array. 10.3.1 New material combinations From theoretical considerations and experimental results, there are indications suggesting that hydrogen terminationdand, thus, OH group formationdis energetically unfavorable on most magnesium oxide (MgO) surfaces.79,85 It has also been postulated that hydrogen adsorption at the insulator/metal interface of the MgO/Pt system would occur on the metal, rather than on the insulator side of the interface. Experimental results point 328 M. Andersson et al. in the same direction, showing that field effect sensors based on MgO/Pt structures exhibit no or very little response/sensitivity to hydrogen. The very small hydrogen-induced response is also in the opposite direction to the normal hydrogen response of SiO2/Pt structures, as briefly discussed earlier, indicating hydrogen adsorption to the metal side of the interface.56 Furthermore, sensitivity to CO of devices comprising dense Pt gate films on top of MgO is extremely low or nonexistent, providing further indications for the response to CO of SiO2/Pt-based sensors at least partly being mediated by an increase in sensitivity toward background hydrogen. 10.3.2 Tailor-made sensing layers for oxygen With the introduction of MgO as the top part of the insulating layer in field effect sensors, the cross-sensitivity to hydrogen or substances containing hydrogen can thus be markedly reduced. This has also been shown for field effect devices with other gate contacts than Pt. By using conducting oxides as gate materialdsuch as iridium oxide (IrO2) or ruthenium oxide (RuO2), for which the work function changes as a function of oxidation state86dit has been shown that the sensitivity toward oxygen, and thereby the gas- sensing abilities of field effect sensors, can also be retained when MgO is used as the insulator87 (see Fig. 10.6). This realizes oxygen sensors with

Senso signal (V) (a) 0,30 1 51015 20% O 0,25 2 2

0,20

0,15 0,10 Senso signal (V) (b) 0,05 0,40 0,00 0,35 0 120 180 30 60 90 150 0,30 Time (min) 0,25 0,20 0,15 0,10 0,05 0,00 0 30 60 90 Time (min)

Figure 10.6 This ﬁgure displays, in (a), the response of a Pt/IrO2/MgO sensor toward 1%, 2%, 5%, 10%, 15%, and 20% O2 in a background of 0.1% O2 in N2 at 500 C. In (b), the response to 200, 500, and 1000 ppm CO and 100, 250, and 500 ppm propene for the same sensor is given. Reprinted with permission from the Proceedings of the IEEE International Conference on Sensors, Christchurch, New Zealand, October 2009. p. 2031e5. ©2009 IEEE (Andersson et al., 2009). Recent progress in silicon carbide ﬁeld effect gas sensors 329 no need for reference gas, unlike the lambda sensor (in the United States, universal heated exhaust gas oxygen, UHEGO).88,89 Partial oxidation or reduction on exposure to different oxygen concentrations at elevated temperatures changes the work function of the gate material at the gate material/insulator boundary and thereby, as discussed in Section 10.2, also the C/V or I/V characteristics of the device. Similar sensors employing ruthenium oxide nanoparticles deposited on SiO2 as gate material, on the other hand, exhibit more or less the same response characteristics to hydrogen and substances containing hydrogen as Pt/ 90 SiO2 and Ru/SiO2 structures.

10.3.3 Tailoring layers for CO2 and NOx Not only cross-sensitivity issues have been addressed in the development of new sensing materials and material combinations but also possible solutions for the detection of substances (e.g., CO2 and NO2), which have not been possible to detect with the ﬁeld effect sensors developed so far, have been investigated. Ion-conducting materials sandwiched between a porous metal gate contact and the insulator have been studied since 2000.91 On exposure of such structures to the target gas (e.g., O2), the target gas adsorbs on the metal gate surface, picking up charges from the metal and thereby forming e the corresponding ions (e.g., by formation of oxygen anions O ). At the three phase boundaries between the metal, ion conductor, and gas phase, these ions spillover to, and can be incorporated in, the material at vacant positions. Most often, but not always, the material is partly composed by the same atoms/ions as the target gas for detection. In the case of oxygen, the ionic conductor is normally an oxide, such as zirconium oxide (ZrO2), commonly doped by another element, e.g., yttrium, to create more oxygen vacancies.89 At elevated temperatures, the ions start to become mobile, moving through the material from high to low concentration by diffusion through vacancies. As a result, charges are introduced into the electronically nonconducting material and to the interface between the ion-conducting and -insulating layers, thereby, as described in Section 10.2, changing the C/V or I/V characteristics of the device. This diffusion is counteracted by the drift due to the electrical ﬁeld created between the interface and the gate electrode, the latter being held at a constant potential. At equilibrium, the net ion current is zero and the potential drop across the ion-conducting layer, DV, in simple terms is theoretically given by the Nernst relation (Eq. (10.5)): 330 M. Andersson et al.

½A ¼ ð Þþ RT : surface DV Voffset T In ½ (10.5) nF A interface where Voffset is the potential difference inherent to the material or materials combination, R is the molar gas constant, T is the temperature, n is the number of electrons transferred per reaction, F is the Faraday constant, and [A]surface and [A]interface are the concentrations of species A at the surface and interface, respectively. For details, see, for instance.92,93 In conjunction with SiC-based field effect sensors, this concept was mainly used earlier for the detection of oxygen. Examples of ion-conducting materials introduced between the porous gate contact and the insulator include 94,95 96 ZrO2, cerium dioxide (CeO2), and perovskite compounds such as 97,98 barium tin oxide (BaSnO3). ZrO2 has been shown to work fairly well for oxygen assessments at high temperatures (600 C and above), whereas devices based on the ionic conductor lanthanum trifluoride (LaF3)exhibita good oxygen sensitivity at lower temperatures, down to room tempera- 99,100 ture. Furthermore, the combination of MgO (as insulator) and LaF3 in field effect devices has also given indications on the possibility for oxygen sensing with markedly reduced cross-sensitivity to hydrogen or substances containing hydrogen. Possibilities for oxygen sensing at lower operating temperatures, as compared with the abovementioned MgO/conducting oxide combinations, have thus been demonstrated.87 Furthermore, the concept has also been utilized for the development of field effect devices sensitive to carbon dioxide (CO2) and nitrogen dioxide (NO2). A solid electrolyte was introduced, which facilitated incorporation 2 of carbonates (CO3 ) or nitrites/nitrates (NO2 /NO3 )inthe ion-conducting layer. In combination with a catalytic gate electrode, from which the carbonates/nitrates are generated on adsorption of CO2/NO2 (through the reaction with adsorbed oxygen anions), field effect sensor devices based on the same principles as oxygen sensors can be realized.101,102 In analogy with the oxygen sensors, the transfer of charge, in the form of 2 CO3 or NO3 , from the gate electrode into the electronically nonconducting solid electrolyte, gives rise to a change in the C/V or I/V characteristics of the device. Earlier investigations have also shown promising results regarding NO2 sensing with Si-based MIS capacitor structures employing 101 sodium nitrite (NaNO2) as the ion-conducting material. For detection of CO2 with a carbonate-based electrolyte, device operating temperatures of at least 400 C are required. Therefore, SiC-based devices are chosen, see Section 10.1, for the development of a SiC field Recent progress in silicon carbide field effect gas sensors 331

effect CO2 sensor based on the binary lithium carbonate (Li2CO3)/barium carbonate (BaCO3) solid electrolyte (see Fig. 10.7). The binary ion conductor exhibits, in addition to good sensitivity to CO2, an excellent stability also under humid conditions. Fig. 10.7 shows a device with electrolyte deposited on top of MgO and a highly porous Pt gate electrode with 103 promising results regarding CO2 monitoring. In this case, MgO also acts as a passivation layer, preventing lithium ions (Liþ) from diffusing into the insulating layer during processing and operation of the device. Perovskites are used as NOx storage materials in catalytic converters for 104 diesel engine exhaust after treatment. Strontium titanate (SrTiO3) has been employed as gate material in SiC-FET devices for NOx detection. Single digit ppm detection was demonstrated between 550 and 600 C, while at lower temperature, i.e., 530 C, the response to NOx was some- 105 what lower but compensated by improved selectivity to NH3.

(a) (b) T = 400°C Auxiliary layer C = 337 pF

Porous Pt sensing Pt(400 nm) Bonding pad electrode (20nm) Ti(10 nm) Air MgO(100 nm) Air SiOx(25 nm)

Si3N4(25 nm) 1% SiO (50 nm) 2 10% μ n-SiC epi (5 m) LCR meter 20%

4H-SiC (n-type) 100 mV

20 min Ni(100 nm) Backside (c) Ti(10 nm) ohmic contact Pt(400 nm) 600 Air base

550 ) mV G n = 1.5 500

450 Sensor signals (V

400 1 10 102 CO2 concentration/%

Figure 10.7 In (a), a schematic drawing of the MISiC ﬁeld effectebased CO2 sensor with a binary carbonate (Li2CO3-BaCO3) auxiliary layer is given; (b) and (c) show the response characteristics during CO2 exposure at 400 C for the device in (a). 332 M. Andersson et al.

10.4 Dynamic sensor operation and advanced data evaluation To improve selectivity toward certain gaseous substances for which detection, discrimination, and quantification otherwise might be difficult due to interference from other gases, the remedy has often been the introduction of more sensors, each with its own cross-sensitivity pattern. Normally, the combination of sensors and sensitivity patterns is very complicated, involving a large number of different kinds of sensors60 or similar sensors operated at different temperatures. The large number of sensor signals and their individual cross-sensitivities make necessary to reduce dimensionality by using multivariate statistical data analysis and pattern recognition methods to retrieve the desired information. The most common method to reduce dimensionality is principal component analysis (PCA).106,107 Multivariate methods such as PCA have, for example, been used in conjunction with SiC-based field effect devices to monitor the combustion process in biomass fueled power plants13 for the estimation of ammonia concentration in typical flue gases65 and for fast lambda control of a gasoline engine.108 Another example of a multivariate analysis method is linear discriminant analysis (LDA).109,110 In analogy with PCA, new variables (discriminant functions) are introduced as linear combinations of the original variables. Whereas PCA is an unsupervised method, in LDA the assignment of sensor observations into predefined groupsee.g., corresponding to concentrations of a certain target gaseis a prerequisite already when constructing the new variables. The linear combinations of sensor signals are calculated such that the distances between the centers of predefined groups are maximized in the new projected data set, while minimizing the scatter among observations within the different groups. This makes LDA a supervised method. As was discussed earlier, the interactions between a certain gate material and the substances of the surrounding gas matrix are temperature- dependent. Different substances show different temperature dependence, which is the reason why operation of a sensor at different temperatures can provide more information about the gas matrix composition, or the concentration of a specific gas in a background of other gases. Instead of an array of sensors, each of them operated at a different temperature, the operation of one sensor in a cycled temperature operation mode can provide just as much, or even more, information. In this way, not only the application of more temperatures is simplified but there is also the benefit of automatically obtaining Recent progress in silicon carbide field effect gas sensors 333 information from nonequilibrium conditions, when changing from one temperature to another, aiding in the discrimination between gases and concentrations. The mean value of the sensor signal at different temperatures, as well as the derivatives of the signal corresponding to temperature changes, can then be extracted and treated by multivariate statistical methods (just as for the case of signals from many individual sensors). Another advantage related to the use of one sensor as a virtual sensor array includes a reduction of drift problems and, overall, a better control of the sensor signal and its stability over time. This approach has been developed using commercial resistive-type MOS sensorsdfor example, for early fire detection in coal mines.111,112 This concept is now also applied to field effect sensor devices based on SiC for detection and quantification of NO2, SO2, discrimination between different gases (such as H2,NH3, and CO), and different concentrations for both Pt and Ir gate field effect sensors.113,114,115 It was also possible to discriminate three different volatile organic compound (VOC) molecules, formaldehyde (50, 100, 150 ppb), benzene (1, 3, 5 ppb), and naphthalene (5, 20, 35 ppb) from each other in a mixture of them in humid air using an Ir-gated SiC- FET and a 1-min temperature cycle.19 Gate bias ramping of SiC-FET devices introduced hysteresis in the sensor signal, the shape of which revealed more information about the gas mixture under testing. Therefore, gate bias cycling is another alternative for dynamic mode operation. The interaction between the various gaseous substances and the gate material is not only dependent on their identity and temperature but also on the gate potential. Temperature cycled operation combined with gate bias cycling improved the resolution when discriminating and quantifying NO2, CO, and 116 NH3. Mixtures of four gases (NH3,CO,NO,andCH4) at two different concentrations (250 and 500 ppm) could be discriminated by employing LDA evaluation.117 Apart from the ambient condition, the shape of the hysteresis varied also with rate of the bias sweep and, of course, the temperature. This was assumed to show the existence of at least two competing chemical processes taking place on the sensor surface, which are also sensitive to the level of the applied gate bias. Fig. 10.8 shows an example of combined temperature and bias cycled operation (TCO-GBCO), feature extraction, and discrimination of NH3 and CO in a background of dry N2. Bastuck et al. investigated the complementary effects from using both MOS sensor devices and SiC-FET sensors in advanced operation modes. The MOS sensors were used in TCO mode, with and without a preconcentrator system, and the SiC-FET sensors were operated in a combined temperature cycled operationegate bias cycled operation (TCO-GBCO) mode.118 334 M. Andersson et al.

(a) (b) 6 320 CO (500 ppm) (300°C) 5 310 400 T 4 set 300 VGS 3 T 290 real 300 2 280 A) μ (V) (

GS 1 270 OS V I 200 0 260 Temperature (°C) –1 250

–2 240 100 0% O 0% r.h., 777 mV/s –3 230 2– 0 10 20 30 40 50 60 –2 –1 0 1 2 3 4 5 Time (s) VGS (V) (c) Dry nitrogen 250/300°C 3 NH3 (500 ppm) CO (500 ppm) 2

–1

–2

–3 Second discriminant function (0.11 %) 0% O2–0% r.h., 777 mV/s –30 –20 –10 0 10 First discriminant function (99.89%) Figure 10.8 (a) TCO-GBCO combined cycle. (b) Features possible to extract from a hysteresis curve: maximum horizontal and vertical width (red arrows), maxima positions (red dots), and the enclosed area by the curve (light red). (c) Linear discriminant analysis discrimination between the dry N2 background, NH3, and CO. The hysteresis features used are horizontal and vertical width and area of ramps from 2 to 5 V, i.e., 777 mVs1 (9 s) at 250 and 300 C together with the signal mean value of each cycle (over 0e10 s).117

The development of two different sensor operating modes may also open- up possibilities regarding self-diagnostic sensor systems. Comparison of the data from two independent methodsde.g., temperature and bias cyclingdmay increase the chances for fault detection and self-diagnosis of the sensor. In the event of a sensor malfunction, it is not likely that the outcome of two separate evaluation schemes would be similar, the discrepancy between them therefore indicating problems. The concept has been demonstrated for a resistive-type metal oxide, MOX, semiconductor sensor utilizing simultaneous temperature cycling, and electrical impedance spectroscopy measurements.119 Recent progress in silicon carbide ﬁeld effect gas sensors 335

10.5 Applications Except for long-term stable sensors, ﬁeld measurements require suitable packaging of the sensors and functional electronics. In the following section, we will review important improvement in the packaging of SiC-FET gas sensors. 10.5.1 Sensor packaging The transistor outline (TO) header is an industrial standard that has been used for several decades to provide a mechanical basis for the installation of electronic and optical components such as semiconductors and laser diodes, while at the same time providing power to the components with the aid of pins. The SiC-FET gas sensor research improved considerably when TO headers were introduced for microelectronic packaging applications. Fig. 10.9 shows a SiC-FET sensor device mounted on a ceramic (Al2O3)substrate,withathin resistive-type Pt heater wire on the backside, together with a Pt100 temperature sensor. The leads of the heater substrate and temperature sensor are spot welded to a gold-plated 16-pin TO8 header, whereas the sensors’ electrical contacts are connected to the pins of the TO8 header by gold wire bonding. Such sensor packaging enables operation temperatures even above 600 C, with good control of temperature and data acquisition. As an example of high temperature applications, TO headers have been used in engine exhaust systems and in ﬂue gas channels in bioheaters. Over the last years, an innovative packaging technology based on LTCC has been developed, improving the performance of the SiC-FET sensors and widening the range of possible applications, see Fig. 10.9 (top right). This technology is characterized by her- metically sealed modules processed from sheets of unsintered LTCC, which

Figure 10.9 Four-inch diameter SiC wafer with about 2000 sensor chips commercially processed. Close-up of the wafer shows single transistor devices. After dicing the chips are mounted in a ceramic package (top right) or in a 16-pin TO8 header (bottom right). 336 M. Andersson et al. are provided by cavities and vias by laser cutting and electrical contact by screen printing. The sheets are then stacked and finally sintered in an oven at 850 C, which renders a ceramic component.120 Nowak et al. presented LTCC packaging of a SiC-based hydrogen sensor, which is glued on the screen-printed contacts.121 Sobocinski et al. demonstrated for the first time a SiC-FET sensor chip introduced in the LTCC stack and cofired in one single step to a packaged device, which do not need any glue or bond wires.122 This requires LTCC sheets, which do not shrink in the x-y direction (Hereaus Gmbh) during sintering.123 The electrical and sensing properties of the SiC-FET gas sensor are retained after the sintering process at 850 C.124 10.5.2 Applications and field tests The outstanding properties, e.g., in terms of long-term stability and high temperature performance of the SiC material in gas sensor devices are manifested in a range of successful applications and field tests. Loloee et al. demonstrated the robustness of SiC-based sensors using the same Pt-SiO2-SiC capacitive devices for continuous hydrogen monitoring in a coal gasification plant during 5 days and, after that, during 20 days in the laboratory.125 One SiC transistor device has also been operated in a small bioheater for more than 42 months. Control of the inlet air to the bioheater by two SiC-FET gas sensors and a temperature sensor increased the efficiency of the combustion of the wood fuel and considerably decreased the emissions of CO and hydrocarbons.7,14 Successful monitoring of ammonia in the exhausts of a diesel engine equipped with selective catalytic reduction system was demonstrated already 200516 and the sensors were successfully tested in two diesel trucks.4 Not only emissions from vehicles and industrial plants are a threat to our health, even indoor environments in private and public buildings need to be controlled. Indoor air pollution is one of the top five environmental risks to public health which significantly affect quality of life and economy. The list of top-10 gases in air pollution includes the so-called VOCs, a wide class of carbonehydrogen-containing chemicals which are normally found in many products of common use, e.g., tobacco smoke, paints, detergents, glues, construction materials, and pressed-wood products. In 2010, the World Health Organization (WHO) released guidelines for a range of hazardous VOCs, e.g., formaldehyde, benzene, and naphthalene, which are frequently found in indoor environments in concentrations of health concern. Formaldehyde, regarded as the most prevalent VOC, is classified as a probable human carcinogen with a recommended exposure limit of 81 ppb during 30 min of exposure. Benzene is classified as a known human carcinogen at any level of exposure. Naphthalene is reported as carcinogenic in animal experiments Recent progress in silicon carbide field effect gas sensors 337 and a possible human carcinogen, the exposure limit for this substance is set to 1.9 ppb as an average annual level.126 Developing sensor systems specifically selective for such gases at the low ppb or even sub-ppb levels has become a market demand priority. Recently, it was demonstrated that the SiC-FET devices detect formaldehyde and naphthalene at concentrations below the recommended exposure limits, i.e., 10 ppb CH2O and below 0.5 ppb C10H8 under 60% RH. Moreover, the SiC-FET device has proven to detect benzene down to 0.2 ppb in 20% RH and 1e3 ppb in 60% RH, see also Fig. 10.3.8,18 In the last couple of years, different field test campaigns have been carried out in the framework of local collaborations or European projects (e.g., SENSIndoor, Key-VOCs). As an example, an experiment at an elementary school was carried out during a period of 3 months for specific detection of formaldehyde. A commercial formaldehyde monitor (FM-801, Graywolf) and a carbon dioxide concentration, temperature, and relative humidity transmitter (tSense Touch Screen CO2 þ RH/T Transmitter, SenseAir AB) were used as reference instruments. The FM-801 formaldehyde meter provides a measurement range from 20 ppb to 1 ppm and records one data point every 30 min. By using the SiC-FET-based sensor system in dynamic operation mode, continuous monitoring is significantly improved allowing data point recording approximately every minute (depending on the temperature cycle used). Fig. 10.10 displays a temperature cycle of 80 s (four temperatures) and extraction of virtual sensor signals in the school tests.20

Figure 10.10 Temperature cycle (blue, solid line), temperature (blue, dashed line), and the raw sensor signal (black line) during one cycle. The mean of four different areas is computed and the middle of each one, marked by a colored dot, can be regarded as a virtual sensor (left panel). In the right panel, the data of the virtual sensors is extracted from the raw sensor signal. 338 M. Andersson et al.

CO2 conc r.h.

Amb. temp. Form. conc.

PLSR1

Ventilation

Aug 20 Aug 21 Aug 22 Aug 23 Aug 24 Aug 25 Aug 26 Aug 27 Aug 28 Aug 29 Aug 30 2016 Figure 10.11 Sensors signals from reference instruments from a period of 11 days, standardized and shifted for visualization. The virtual sensors have been used to build the PLSR1 signal (red), smoothed with a window size of 22 (w30 min) for visualization. The bottom blue line represents the normal schedule of the ventilation of the school. The start of each day, i.e., midnight, is marked on the x-axis, and night from 6 p.m. to 6 a.m. is marked by darker areas. Note that Aug 20/21 and Aug 27/28 there is no school (weekends).

In Fig. 10.11, measurements from 11 days are shown. Using a multivariate regression model based on partial least squares regression (PLSR) on the sensor data, the experiment demonstrated a very good correlation between the SiC- FET sensor and the FM-801 meter. The formaldehyde builds up at night and during weekends while the ventilation is switched off. The highest peak for the reference instrument was 34 ppb (August 21), while the computed PLSR1 signal from the SiC-FET sensor data had a peak of 24.5 ppb (August 28). In summary, the formaldehyde always stayed well below the threshold value of 80 ppb. The data evaluation also revealed some possible cross- sensitivity of the SiC-FET sensor to other common VOCs that are emitted by breath (e.g., acetone and isoprene), which is an area to be further investigated.20

10.6 Summary The SiC-FET devices as high-temperature gas sensors are commercially available in sensor systems for combustion control, e.g., in small- and medium-scale power plants. Research and development has realized tailor-made sensing layers for, e.g., oxygen and carbon dioxide detection. Detection of toxic indoor gases, VOCs, below legally restricted levels has been demonstrated. Temperature and bias cycled operation modes together Recent progress in silicon carbide field effect gas sensors 339 with advanced data evaluation based on multivariate statistics improved selectivity and sensitivity in complex gas mixtures. Field test campaigns have demonstrated the suitability of using the SiC-FET sensor as a selective formaldehyde sensor. Acknowledgments Grants are acknowledged from the VINN Excellence Center in research and innovation on Functional Nanoscale Materials (FunMat), the Swedish Governmental Agency for Innova- tion Systems (VINNOVA #621-2012-4497), and the Swedish Research Council (VR #621-2012-4497). The authors also acknowledge funding from the European Union’s Sev- enth Programme for research, technological development, and demonstration under grant agreement No. 604311 (SENSIndoor), and from the COST Action TD1105 (EuNetAir). A.L.S. acknowledges the Swedish Government Strategic Research Area in Materials Science on Functional Materials at Linkoping€ University (Faculty Grant SFO-Mat-LiU No. 2009- 00971). Dr Ruth Pearce is acknowledged for the contribution in the first edition of this book chapter.40 The epitaxial graphene sensor area has grown into an independent research area, exemplified in the introduction. References 1. Bergveld P. Development of an ion sensitive solid state device for neuro-physiological measurements. IEEE Trans Biomed Eng 1970;17:70e1. 2. Lundstrom€ I, Shivaraman S, Svensson C, Lundkvist L. A hydrogen-sensitive MOS field-effect transistor. Appl Pys Lett 1975;26(2):55e7. 3. Lundstrom€ KI, Shivaraman MS, Svensson CM. A hydrogen-sensitive Pd-gate MOS transistor. J Appl Phys 1975;46:3876e81. 4. Lundstrom€ I, Sundgren H, Winquist F, Eriksson M, Krantz-Rulcker€ C, Lloyd- Spetz A. Twenty-five years of field effect gas sensor research in Linkoping.€ Sensor Actu- ator B 2007;121:247e62. 5. Trinchi A, Kandasamy S, Wlodarski W. High temperature field effect hydrogen and hydrocarbon gas sensors based on SiC MOS devices. Sensor Actuator B Chem 2008;133: 705e16. 6. Lloyd Spetz A, Skoglundh M, Ojam€ae L. FET gas-sensing mechanism, experimental and theoretical studies. ch. 4. In: Comini E, Faglia G, Sberveglieri G, editors. Solid state gas sensing, New York. Norwell MA, USA: Springer; 2009, ISBN 978-0-387-09664-3. p. 153e79. 7. Andersson M, Pearce R, Lloyd Spetz A. New generation SiC based field effect transistor gas sensors. Sensor Actuator B Chem 2013;179:95e106. https://doi.org/10.1016/ j.snb.2012.12.059. 8. Puglisi D, Eriksson J, Andersson M, Huotari J, Bastuck M, Bur C, Lappalainen J, Schuetze A, Lloyd Spetz A. Exploring the gas sensing performance of catalytic metal/metal oxide 4H-SiC field effect transistors. Mater Sci Forum 2016;858: 997e1000. 9. Hunter GW, Neudeck P, Jefferson GD, Madzar GC, Liu CC, Wu QH. The development of hydrogen sensor technology at NASA Lewis research center. In: Proceedings of the 4th annual space system health management technology conference, Cincinatti, USA; 1992. 340 M. Andersson et al.

10. Spetz A, Arbab A, Lundstrom€ I. Gas sensors for high temperature operation based on metal oxide silicon carbide (MOSiC) devices. In: Proceedings of Eurosensors VI, San Sebastian, Spain; 1992. p. 19e23. 11. Svenningstorp H, Unéus L, Tobias P, Lundstrom€ I, Ekedahl L-G, Lloyd Spetz A. High temperature gas sensors based on catalytic metal field effect transistors. Mater Sci Forum 2000;338e342:1435e8. 12. Wingbrant H, Svenningstorp H, Salomonsson P, Tengstrom€ P, Lundstrom€ I, Lloyd Spetz A. Using a MISiCFET device as a cold start sensor. Sensor Actuator B 2003;93: 295e303. 13. Unéus L, Artursson T, Mattsson M, Ljung P, Wigren R, Mârtensson P, Holmberg M, Lundstrom€ I, Lloyd Spetz A. Evaluation of on-line flue gas measurements by MISiC- FET and metal oxide sensors in boilers. IEEE Sens J 2005;5(1):75e81. 14. Andersson M, Everbrand L, Lloyd Spetz A, Nystrom€ T, Nilsson M, GauffinC, Svensson H. A MISiCFET based gas sensor system for combustion control in small- scale wood fired boilers. Proc IEEE Sensors 2007:962e5. Atlanta, USA Oct 28e31. 15. Andersson M. SiC based field effect sensors e from basic understanding to tailoring of devices for high temperature applications IX International workshop on semiconductor gas sensors. SGS 2015. 13e16 December, Zakopane, Poland, 2015. 16. Wingbrant H, Svenningstorp H, Salomonsson P, Kubinski D, Visser JH, Lofdahl€ M, Lloyd Spetz A. Using a MISiC-FET sensor for detecting NH3 in SCR systems. IEEE Sens J 2005;5(5):1099e105. 17. Andersson M, Wingbrant H, Petersson H, Unéus L, Svenningstorp H, Lofdahl€ M, Holmberg M, Lloyd Spetz A. Gas sensor arrays for combustion control. In: Ranch S, editor. Encyclopedia of sensors, vol. 4. USA: American Scientific Publishers; 2006. p. 139e54. 18. Puglisi D, Eriksson J, Bur C, Schuetze A, Lloyd Spetz A, Andersson M. Catalytic metal-gate field effect transistors based on SiC for indoor air quality control. J Sens Sens Syst (JSSS) 2015;4:1e8. 19. Bur C, Bastuck M, Puglisi D, Schutze€ A, Lloyd Spetz A, Andersson M. Discrimination and quantification of volatile organic compunds in the ppb-range with gas sensitive SiC-FETs using multivariate statistics. Sensor Actuator B Chem 2015;214:225e33. 20. Bastuck M, Puglisi D, Moller€ M, Reimringer W, Schutze€ A, Lloyd Spetz A, Andersson M. Low-cost chemical gas sensors for selective formaldehyde quantification at ppb-level in the field. In: Proc. AMA Conf. 2017 sensor 2017 and IRS2; 2017. p. 702e7. 21. Schalwig J, Ahlers S, Kreisl P, Bosch v, Braunmuhl€ C, Muller€ G. A solid-state gas sensor array for monitoring NOx storage catalytic converters. Sensor Actuator B 2004;101:63e71. 22. Casals O, Becker T, Godignon P, Romano-Rodriguez A. SiC-based MIS gas sensor for high water vapor environments. Sensor Actuator B Chem 2012;175:60e6. 23. Gurbuz Y, Kang WP, Davidson JL, Kerns DV. High-temperature tolerant diamond diode for carbon monoxide gas detection. J Appl Phys 1998;84:6935e6. 24. Schalwig J, Kreisl P, Ahlers S, Muller€ G. Response mechanism of SiC-based MOS field-effect gas sensors. IEEE Sens J 2002;2(5):394e402. 25. Eickhoff M, Schalwig J, Weidemann O, Gorgens€ L, Neuberger R, Hermann M, Steinhoff G, Baur B, Muller€ G, Ambacher O, Stutzmann M. Electronics and sensors based on pyroelectric Al-GaN/GaN heterostructures. Phys Status Solidi C: Conf 2003; 0(6):1908e18. 26. Chen H-I, Cheng C-H, Chen W-C, Liu I-P, Lin K-W, Liu W-C. Hydrogen sensing performance of a Pd nanoparticle/Pd film/GaN-based diode. Sensor Actuator B Chem 2017;247:514e9. Recent progress in silicon carbide field effect gas sensors 341

27. Chou P-C, Chen H-I, Liu I-P, Chen W-C, Chen C-C, Liou J-K, Lai C-J, Liu W-C. On a Schottky diode-type hydrogen sensor with pyramid-like Pd nanostructures. Int J of Hydrogen Energy 2015;40:9006e12. 28. Guo Y, Wang X, Miao B, Li Y, Yao W, Xie Y, Li J, Wu D, Pei R. An AuNPs- functionalized AlGaN/GaN high electron mobility transistor sensor for ultrasensitive detection of TNT. RSC Adv 2015;5:98724e9. 29. Offermans P, Si-Ali A, Brom-Verheyden G, Greens K, Lenci S, van Hove M, Decoutere S, Van Schaijk R. Suspended AlGaN/GaN membrane devices with recessed open gate areas for ultra-low-power ari quality monitoring. Conf Proc IEEE Sensor 2015;15:871e4. IEDM. 30. Weng MH, Mahapatra R, Wright NG, Horsfall AB. Role of oxygen in high temperature hydrogen sulfide detection using MISiC sensors. Meas Sci Technol 2008;19: 024002 (5 pp.). 31. Davydov AA, Marshneva VI, Shepotko ML. Metal oxides in hydrogen sulphide oxidation by oxygen and sulphur dioxide I. The comparison study of the catalytic activity. Mechanism aof the interactions between H2S and SO2 on some oxides. Appl Catal A 2003;244:93e100. 32. Nakagomi S, Ikeda M, Kokubun Y. Comparison of hydrogen sensing properties of Schottky diodes based on SiC and b-Ga2O3 single crystal. Sens Lett 2011;9(2):616e20. 33. Nakagomi S, Sai T, Kokubun Y. Hydrogen gas sensor with self temperature compensation based on b-Ga2O3 thin film. Sensor Actuator B 2013;187:413e9. 34. Nakagomi S, Kubo S, Kokubun Y. The orientational relationship between monoclinic b-Ga2O3 and cubic NiO. J Cryst Growth 2016;445:73e7. 35. Yakimova R, Virojanadara C, Gogova D, Syv€aj€arvi M, Siche D, Larsson K, Johansson LI. Analysis of the formation conditions for large area epitaxial graphene on SiC substrates. Mater Sci Forum 2010;645e648:565e8. 36. Pearce R, Yakimova R, Eriksson J, Hultman L, Andersson M, Lloyd Spetz A. Development of FETs and resistive devices based on epitaxially grown single layer graphene on SiC for highly sensitive gas detection. Mater Sci Forum 2012;717e720: 680e90. 37. Pearce R, Eriksson J, Iakimov T, Hultman L, Lloyd Spetz A, Yakimova R. On the differing sensitivity to chemical gating of single and double layer epitaxial graphene explored using scanning kelvin probe microscopy. ACS Nano 2013;7(5):4647e56. 38. Eriksson J, Pearce R, Iakimov T, Vironajadara C, Gogova D, Andersson M, Syv€aj€arvi M, Lloyd Spetz A, Yakimova R. The influence of substrate morphology on thickness uniformity and unintentional doping of epitaxial graphene on SiC. Appl Phys Lett 2012;100:241607. 39. Eriksson J, Puglisi D, Vasiliauskas R, Lloyd Spetz A, Yakimova R. Thickness uniformity and electron doping in epitaxial graphene on SiC. Mater Sci Forum 2013; 740e742:153e6. 40. Andersson M, Lloyd Spetz A, Pearce R. Recent trends in silicon carbide (SiC) and graphene based gas sensors. In: Janiso R, Tan OK, editors. Semiconductor gas sensors. 1st ed. Philadelphia: Woodhead publishing; 2013. p. 117e58. 41. Eriksson J, Puglisi D, Kang Y-H, Yakimova R, Lloyd Spetz A. Adjusting the electronic properties and gas reactivity of epitaxial graphene by thin suface metallization. Physica B 2014;439:105e8. 42. Eriksson J, Puglisi D, Strandqvist C, Gunnarsson R, Ekeroth S, Ivanov IG, Helmersson U, Uvdal K, Yakimova R, Lloyd Spetz A. Modified epitaxial graphene on SiC for extremely sensitive and selective gas sensors. Mater Sci Forum 2016;858: 1145e8. 342 M. Andersson et al.

43. Rodner M, Puglisi D, Yakimova R, Eriksson J. A platform for extremely sensitive gas sensing: 2D materials on silicon carbide. In: TechConnect briefs. Materials for energy, efficiency and sustainability, 2; 2018, ISBN 978-0-9988782-3-2. p. 101e4. 44. Shtepliuk I, Eriksson J, Khranovskyy V, Iakimov T, Lloyd Spetz A, Yakimova R. Monolayer graphene/SiC Schottky barrier diodes with improved barrier height uniformity as a sensing platform for the detection of heavy metals. Beilstein J Nanotechnol 2016;7:1800e14. 45. Rodner M, Bahonjic J, Mathisen M, Ekeroth S, Helmersson U, Ivanov IG, Yakimova R, Eriksson J. Performance tuning of gas sensors based on epitaxial graphene on silicon carbide. Mater Des 2018;153:153e8. 46. Sze SM. Physics of semiconductor devices. New York: John Wiley & Sons; 1981. 47. Lundstrom€ I. Field effect chemical gas sensors, In: Gopel€ W, Hesse J, Zemel J.N, editors. Sensors A comprehensive survey, 1991;2(1). Wienheim: VCH Verlagsgesellschaft mbH p. 467-528. 48. Dimitrijev S. Understanding semiconductor devices. New York: Oxford University Press; 2000. 49. Bur C. Selectivity enhancement of gas sensitive field effect transistors by dynamic operation. Link€oping Stud Sci Technol 2015:24e31. Dissertation No. 1644. ISBN: 978-91-7519-119-5. 50. Senft C, Galonska T, Widanarto W, Eisele I, Frerichs HP, Wilberts M. Stability of FET-based hydrogen sensors at high temperatures. Proc IEEE Sensors Conf 2007: 189e92. 4388368. 51. Fogelberg J, Eriksson M, Dannetun H, Petersson L-G. Kinetic modeling of hydrogen ad/absorption in thin films on hydrogen sensitive field effect devices: observation of large hydrogen dipoles at the Pd-SiO2 interface. J Appl Phys 1995;78(2):988e96. 52. Ekedahl L-G, Eriksson M, Lundstrom€ I. Hydrogen sensing mechanisms of metal- insulator interfaces. Acc Chem Res 1998;31(5):249. 53. Wallin M, Gronbeck€ H, Lloyd Spetz A, Eriksson M, Skoglundh M. Vibrational analysis of H2 and D2 adsorption on Pt/SiO2. J Phys Chem B 2005;109:9581e8. 54. Eriksson M, Lundstrom€ I, Ekedahl L-G. A model of the Temkin isotherm behavior for hydrogen adsorption at Pd-SiO2 interfaces. J Appl Phys 1997;82(6):3143e6. 55. Usagawa T, Kikuchi Y. Device characteristics for Pt-Ti-O gate Si-MISFETs hydrogen gas sensors. Sensor Actuator B 2011;160(1):105e14. 56. Eriksson M, Ekedahl L-G. Hydrogen adsorption states at the Pd/SiO2 interface and simulation of the response of a Pd metal-oxide-semiconductor hydrogen sensor. J Appl Phys 1998;83(8):3947e51. 57. Schalwig J, Muller€ G, Eickhoff M, Ambacher O, Stutzmann M. Group III-nitride based gas sensors for combustion monitoring. Mater Sci Eng B 2002;93:207e14. 58. Weidemann O, Hermann M, Steinhoff G, Wingbrant H, Lloyd Spetz A, Stutzmann M, Eickhoff M. Influence of surface oxides on hydrogen-sensitive Pd-GaN Schottky diodes. Appl Phys Lett 2003;83(4):773e5. 59. Eriksson M, Salomonsson A, Lundstrom€ I, Briand D, Åbom AE. The influence of the insulator surface properties on the hydrogen response of field-effect gas sensors. J Appl Phys 2005;98(3):34903e8. 60. Roy SK, Furnival BJD, Wood NG, Vassilevski KV, Wright NG, Horsfakk AB, OMalley CJ. SiC gas sensor array for extreme environment. Proc IEEE Sensors 2013:250e3. 978-1-4673-4642-9/13. 61. Ofrim B, Udrea F, Brezeanu G, Hsieh AP-S. Hydrogen sensor base on silicon carbid (SiC) MOS capacitor. Proc IEEE Sensors 2012:367e70. 62. Baranzahi A, Lloyd Spetz A, Glavmo M, Carlsson C, Nytomt J, Salomonsson P, Jobson E, H€aggendal B, Mårtensson P, Lundstrom€ I. Response of metal-oxide-silicon carbide sensors to simulated and real exhaust gases. Sensor Actuator 1997;B43:52e9. Recent progress in silicon carbide field effect gas sensors 343

63. Burch R, Watling TC. Kinetics and mechanism of the reduction of NO by C3H8 over Pt/Al2O3 under lean-burn conditions. J Catal 1997;169:45e54. 64. Burch R, Watling TC. The effect of sulphur on the reduction of NO by C3H6 and C3H8 over Pt/Al2O3 under lean-burn conditions. Appl Catal, B 1998;17:131e9. 65. Andersson M, Ljung P, Mattsson M, Lofdahl€ M, Lloyd Spetz A. Investigations on the possibilities of a MISiCFET sensor system for OBD and combustion control utilizing different catalytic gate materials. Top Catal 2004;30/31:365e8. 66. Kahng YH, Lu W, Tobin RG, Loloee R, Ghosh RN. The role of oxygen in hydrogen sensing by a platinum-gate silicon carbide sensor: an ultrahigh vacuum study. J Appl Phys 2009;105:064511. 67. Chilton TH. The manufacture of nitric acid by the oxidation of ammonia. In: Chemical engineering progress monograph series, vol. 3. New York: American Institute of Chemical Engineers; 1960. 56. 68. Fogelberg J, Lundstrom€ I, Petersson L-G. Ammonia dissociation on oxygen covered palladium studied with a hydrogen sensitive Pd-MOS device. Phys Scripta 1987;35: 702e5. 69. Spetz A, Armgarth M, Lundstrom€ I. Optimization of ammonia-sensitive structures with platinum gates. Sensor Actuator 1987;11:349e65. 70. Lofdahl€ M, Utaiwasin C, Carlsson A, Lundstrom€ I, Eriksson M. Gas response dependence on metal gate morphology of field-effect devices. Sensor Actuator B 2001;80:183e92. 71. Lofdahl€ M. Spatially resolved gas sensing, Link€oping studies in science and technology. Sweden: Linkoping;€ 2001. p. 115e27. Dissertation no. 696. 72. Wallin M, Byberg M, Gronbeck€ H, Skoglundh M, Eriksson M, Lloyd Spetz A. Vibrational analysis of H2 and NH3 adsorption on Pt/SiO2 and Ir/SiO2 model sensors. In: Proceedings of IEEE international conference on sensors, Atlanta, USA; 2007. p. 1315e7. 73. Åbom AE, Haasch RT, Hellgren N, Finnegan N, Hultman L, Eriksson M. Character- ization of the metaleinsulator interface of field-effect chemical sensors. J Appl Phys 2003;93(12):9760e8. 74. Schalwig J, Muller€ G, Karrer U, Eickhoff M, Ambacher O, Stutzmann M, Gorgens€ L, Dollinger G. Hydrogen response mechanism of Pt-GaN Schottky diodes. Appl Phys Lett 2002;80(7):1222e4. 75. Yamaguchi T, Kiwa T, Tsukada K, Yokosawa K. Oxygen interference mechanism of platinum-FET hydrogen gas sensor. Sensor Actuator 2007;136(1):244e8. 76. Dean VW, Frenklach M, Phillips J. Catalytic etching of platinum foils and thin films in hydrogen-oxygen mixtures. J Phys Chem 1988;92:5731e8. 77. Nakagomi S, Tobias P, Baranzahi A, Lundstrom€ I, Mårtensson P, Lloyd Spetz A. Influence of carbon monoxide, water and oxygen on high temperature catalytic metal-oxide silicon carbide structures. Sensor Actuator B 1997;45:183e91. 78. Becker E, Andersson M, Eriksson M, Lloyd Spetz A, Skoglundh M. Study of the sensing mechanism towards carbon monoxide of platinum-based field effect sensors. IEEE Sens J 2011;11(7):1527e34. 79. Darmastuti Z, Pearce R, Lloyd Spetz A, Andersson M. The influence of gate bias and structure on the CO sensing performance of SiC based field effect sensors. Proc IEEE Sensors 2011:133e6. Limerick, Ireland, October 28-31, 2011. 80. Andersson M, Lloyd Spetz A, Pearce R. Tunable gas alarms for high temperature applications based on 4H-SiC MISFET devices. In: Proceedings of the international conference on silicon carbide and related materials, Cleveland, ICSCRM 2011, OH, USA, September 11e16; 2011. p. 365. 81. Andersson M, Lloyd Spetz A. Tailoring of SiC based field effect gas sensors for improved selectivity to non-hydrogen containing species. In: Proc. IMCS13, Perth, Australien 12e14 July; 2010. p. 369. 344 M. Andersson et al.

82. Eriksson M, Ekedahl L-G. The influence of CO on the response of hydrogen sensitive Pd-MOS devices. Sensor Actuator B 1997;42:217e23. 83. Medlin JW, McDaniel AH, Allendorf MD, Bastasz R. Effects of competitive carbon monoxide adsorption on the hydrogen response of metal-insulator-semiconductor sensors: the role of metal film morphology. J Appl Phys 2003;93(4):2267e74. 84. Di Natale C, Buchholt K, Martinelli E, Paolesse R, Pomarico G, D’Amico A, Lundstrom€ I, Lloyd Spetz A. Investigation of quartz microbalance and ChemFET transduction of molecular recognition events in a metalloporphyrin film. Sensor Actu- ator B Chem 2009;135:560e7. 85. Chizallet C, Costentin G, Che M, Delbecq F, Sautet P. Revisiting acido-basicity on the MgO surface by periodic density functional theory calculations: role of surface topology and ion coordination on water dissociation. J Phys Chem B 2006;110: 15878e86. 86. Sang YH, Ho WJ, Jong-Lam L. IrO2 Schottky contact on n-type 4H-SiC. Appl Phys Lett 2003;82(26):4726e8. 87. Andersson M, Lloyd Spetz A. Tailoring of field effect gas sensors for sensing of non-hydrogen containing substances from mechanistic studies on model systems. In: Proceedings of the IEEE international conference on sensors, Christchurch, New Zealand, October; 2009. p. 2031e5. 88. Logothetis EM, Visser JH, Soltis RE, Rimai L. Chemical and physical sensors based on oxygen pumping with solid-state electrochemical cells. Sensor Actuator B 1992;9: 183e9. 89. Visser JH, Soltis RE. Automotive exhaust gas sensing systems. IEEE Trans Instrum Mea- surement 2001;50(6):1543e50. 90. Salomonsson A, Petoral Jr RM, Uvdal K, Aulin C, K€all P-O, Ojam€ae L, Strand M, Sanati M, Lloyd Spetz A. Nanocrystalline ruthenium oxide and ruthenium in sensing applications e an experimental and theoretical study. J Nanoparticle Res 2006;8: 899e910. https://doi.org/10.1007/s11051-005-9058e1. 91. Lloyd Spetz A, Nakagomi S, Wingbrant H, Andersson M, Salomonsson A, Roy S, Wingqvist G, Katardjiev I, Eickhoff M, Uvdal K, Yakimova R. New Materials for Chemical and Biosensors, Mater Manuf Process 2006;21:253e6. 92. Reinhardt G, Mayer R, Rosch€ M. Sensing small molecules with amperometric sensors. Solid State Ionics 2002;150:79e92. 93. Garzon FH, Mukundan R, Lujan R, Brosha EL. Solid state ionic devices for combustion gas sensing. Solid State Ionics 2004;175:487e90. 94. Miyahara Y, Tsukada K, Miyagy H. Field-effect transistor using a solid electrolyte as a new oxygen sensor. J Appl Phys 1987;63(7):2431e4. 95. Tobias P, Macak K, Helmersson U, Lundstrom€ I, Lloyd Spetz A. Zirconia based oxygen sensor without the need of a reference electrode. In: Proceedings of the 8th international meeting on chemical sensors, Basel, Switzerland; 2000. p. 149. 96. Jacobsén S, Helmersson U, Ekedahl L-G, Lundstrom€ I, Mårtensson P, Lloyd Spetz A. Pt/CeO2 SiC Schottky diodes with high response to hydrogen and hydrocarbons. In: Proceedings of transducers ’01 and Eurosensors XV, Munich, Germany; 2001. p. 832e5. 97. Cerda J, Arbiol J, Dezanneau G, Díaz R, Morante JR. Perovskite-type BaSnO3 powders for high temperature gas sensor applications. Sensor Actuator B 2002;84:21e2. 98. Cerda J, Morante JR, Lloyd Spetz A. New tunnel Schottky SiC devices using mixed conduction ceramics. Mater Sci Forum 2003;433(6):949e52. 99. Krause S, Moritz W, Grohmann I. Improved long term stability for an LaF3 based oxygen sensor. Sensor Actuator B 1994;18(19):148e54. 100. Vasiliev A, Moritz W, Fillipov V, Bartholom€aus L, Terentjev A, Gabusjan T. High temperature semiconductor sensor for the detection of fluorine. Sensor Actuator B 1998;49:133e8. Recent progress in silicon carbide field effect gas sensors 345

101. Zamani C, Shimanoe K, Yamazoe N. A new capacitive-type NO2 gas sensor combining an MIS with a solid electrolyte. Sensor Actuator B 2005;109:216e20. 102. Kida T, Kishi S, Yuasa M, Shimanoe K, Yamazoe N. Planar NASICON-based CO2 sensor using BiCuVOx/Perovskite-type oxide as a solid-reference electrode. J Electrochem Soc 2008;155:J117e21. 103. Inoue H, Andersson M, Yuasa M, Kida T, Lloyd Spetz A, Shimanoe K. CO2 sensor combining a metal-insulator silicon carbide (MISiC) capacitor and a binary carbonate. Electrochem Solid State Lett 2010;14(1):J4e7. https://doi.org/10.1149/1.3512998. 104. Abrahamsson B, Gronbeck€ H. NOx adsorption on ATiO3(001) perovskite surfaces. J Phys Chem C 2015;119:18495e503. 105. Moller€ P, Andersson M, Lloyd Spetz A, Puustinen J, Lappalainen J, Eriksson J. NOx sensing with SiC field effect transistors. Mater Sci Forum 2016;858:993e6. 106. Wold S, Sjostr€ om€ M, Eriksson L. PLS-regression: a basic tool of chemometrics. Che- mometr Intell Lab Syst 2001;58:109e30. 107. Chatfield C, Collins AJ. Introduction to multivariate analysis. London: Chapman & Hall; 1980. 108. Larsson O, Goras€ A, Nytomt J, Carlsson C, Lloyd Spetz A, Artursson T, Holmberg M, Lundstrom€ I, Ekedahl L-G, Tobias P. Estimation of air fuel ratio of individual cylinders in SI engines by means of MISiC sensor signals in a linear regression model. In: SAE technical paper series, paper 2002e01e0847, Detroit, Michigan, USA; 2002. 109. Duda RO, Hart PE, Stork DG. Pattern classification. New York: Wiley; 2000. 110. Gutierrez-Osuna R. Pattern analysis for machine olfaction. IEEE Sens J 2002;2(3): 189e202. 111. Lee AP, Reedy BJ. Temperature modulation in semiconductor gas sensing. Sensor Actuator B 1999;60:35e42. 112. Reimann P, Horras S, Schutze€ A. Field-test system for underground fire detection based on semiconductor gas sensors. In: Proceedings of the IEEE international conference on sensors, 2009, Christchurch, New Zealand; 2009. p. 659e64. 113. Bur C, Reimann P, Schutze€ A, Andersson M, Lloyd Spetz A. Increasing the selectivity of Pt-gate SiC field effect gas sensors by dynamic temperature modulation. In: Proc. IEEE int. Conf. on sensors, Waikoloa, USA 1e4 November; 2010. p. 1267e72. https://doi.org/10.1109/ICSENS.2010.5690598. 114. Bur C, Reimann P, Schutze€ A, Andersson M, Lloyd Spetz A. New method for selectivity enhancement of SiC field effect gas sensors for quantification of NOx, microsystem technologies/smart sensors. Actuator MEMS 2012;18(7):1015e25. https://doi.org/10.1007/s00542e012e1434-z. Springer-Verlag, Berlin. 115. Darmastuti Z, Bur C, Lindqvist N, Andersson M, Schutze€ A, Lloyd Spetz A. Hierarchical methods to improve the performance of the SiC e FET as SO2; sensors in flue gas desulphurization systems. Sensor Actuator B 2015;206:609e16. 116. Bur C, Bastuck M, Lloyd Spetz A, Andersson M. Selectivity enhancement of SiC-FET gas sensors by combining temperature and gate bias cycled operation using multivariate statistics. Sensor Actuator B Chem 2014;193:931e40. 117. Bastuck M, Bur C, Lloyd Spetz A, Andersson M, Schutze€ A. Gas identification based on bias induced hysteresis of gas-sensitive SiC field effect transistor. J Sens Sens Syst (JSSS) 2014;3:9e19. 118. Bastuck M, Reimringer W, Conrad T, Schutze€ A. Dynamic multi-sensor operation and read-out for highly selective gas sensor systems. Proc Eng 2016;168:1685e8. 119. Reimann P, Dausend A, Schutze€ A. A self-monitoring and self-diagnosis strategy for semiconductor gas sensor systems. In: Proceedings of IEEE international conference on sensors. Italy: Lecce; 2008. 192e5. 346 M. Andersson et al.

120. Sobocinski M, Zwiertz R, Juuti J, Jantunen H, Golorika L. Electrical and electrome- chanical characteristics of LTCC embedded piezoelectric bulk actuators. Adv Appl Ceram 2010;109(3):135e8. 121. Nowak D, Kulczak D, Januszkiewicz M, Dziedzic A. High temperature LTCC package for SiC-based gas sensor. Opt Appl 2009;XXXIX(4):701e4. 122. Sobocinski M, Khajavizadeh L, Andersson M, Lloyd Spetz A, Juuti J, Jantunen H. Performance of LTCC embedded gas sensors. Procedia Eng 2015;120:253e6. 123. Rabe T, Schiller WA, Hochheimer T, Modes C, Kipka A. Zero shrinkage of LTCC by self-constrained sintering. Int J Appl Ceram Technol 2005;2(5):374e82. 124. Lloyd Spetz A, Sobocinski M, Halonen N, Puglisi D, Juuti J, Jantunen H, Anderson M. LTCC, new packaging approach for toxic gas and particle detection. Proceedia Eng 2015;120:484e7. 125. Loloee R, Chorpening B, Beer S, Ghosh RN. Hydrogen monitoring for power plant applications using SiC sensors. Sensor Actuator B 2008;129:200e10. 126. WHO (World Health Organization). Regional ofﬁce for Europe. WHO guidelines for indoor air quality: selected pollutants, ISBN 978 92 890 0213 4. CHAPTER ELEVEN

Semiconducting direct thermoelectric gas sensors

F. Rettig, R. Moos University of Bayreuth, Bayreuth, Germany

Contents

11.1 Introduction 347 11.1.1 Motivation for research on direct thermoelectric gas sensors 347 11.1.2 Thermoelectric power 349 11.1.3 Direct and indirect thermoelectric gas sensors 350 11.1.4 Early research activities 353 11.2 Direct thermoelectric gas sensors 353 11.2.1 Measurement techniques 353 11.2.2 Modeling and simulation of thermoelectric gas sensors 357 11.2.3 Measurements and results 367 11.2.4 Ionic direct thermoelectric gas sensors 378 11.3 Conclusion and future trends 380 References 381

11.1 Introduction This introductory section will describe the various issues that are motivating research on direct thermoelectric gas sensors (DTEGs) before presenting a brief introduction to thermoelectric power for the general reader. The principles of direct and indirect thermoelectric gas sensors are then outlined, while early research work is reviewed in the ﬁnal subsection. 11.1.1 Motivation for research on direct thermoelectric gas sensors Gas sensors play an important role in many applications and have been extensively developed during the past few decades. This is especially the case for applications in monitoring automotive exhaust gases (lambda probe) and air quality (AQ sensors). Although the lambda probe itself cannot reduce polluting emissions from automobiles, it allows the adjustment of a

Semiconductor Gas Sensors, Second Edition ISBN: 978-0-08-102559-8 © 2020 Elsevier Ltd. https://doi.org/10.1016/B978-0-08-102559-8.00011-2 All rights reserved. 347 j 348 F. Rettig and R. Moos stoichiometric mixture of air and fuel.1 A modern concept with two lambda probes even allows detection of a defect in a three-way catalyst.2 AQ sensors can monitor the air quality in houses and cars,3 as well as detecting concentrations of unburnt hydrocarbons,4 an important point in fire prevention. Other applications include alerting people when harmful gases are in the ambient atmosphere.5 Since the 1960s, many research activities have been addressed to resistive (also known as “conductometric”) gas sensors. Since the development of 6 Taguchi’s sensor based on SnO2, many semiconducting materials have been investigated and analyzed. Besides SnO2, the most prominent exam- 7,8 9,10 11,66 12 13 ples are TiO2; SrTiO3; SrTixFe1exO3ed; WO3; Ga2O3; 14e16 17 Cr2O3; or ZnO. Many of these materials are discussed elsewhere in this book. The appeal of resistive gas sensors is the relative simplicity of manufacturing resistive sensors combined with an uncomplicated principle for taking measurements. Some of these materials have been tested in automotive exhausts (e.g., 8,9,13,18). However, harsh environments are a challenge for resistive gas sensors, as poisoning or deterioration of the gas sensitive materials by aggressive components such as SO2 or NOx may occur or abrasion of the gas sensitive layer by particle-containing high-velocity gas streams may cause irreversible harm to the gas sensors.19 This is easy to understand because each geometric changedwhich may occur, for instance, by abrasiondcan have a marked effect on the resistance and cause flawed concentration readings. Protective layers were proposed to overcome these problems.20,21 For application in exhausts, potentiometric or amperometric gas sensors based on ion conduction membranes of yttria-stabilized zirconia (YSZ) are typically used.1,22 Such sensors provide sufficient stability against harsh environments. The potentiometric principle offers the possibility of measuring a path-independent quantitydthe electrical potential difference (voltage). In theory, abrasion does not significantly affect the sensor signals. These advantages come at the cost of a more complicated design where potentiometric and amperometric gas sensors are concerneddfor instance, the classical lambda probe requires an air Ref. 23 or a pumped Ref. 24. Semiconducting DTEGs do not have the disadvantages of resistive or potentiometric gas sensors. The measurand is a path-independent thermovoltage and no gas reference is required. The typical materials used in resistive gas sensors can also be utilized for DTEGs. In this chapter, it will be shown that intrinsic semiconducting materials have an enhanced sensitivity compared with classical p- or n-type conducting materials. Semiconducting direct thermoelectric gas sensors 349

These advantages are the main drivers for the research and development of DTEGs. In the next section, a short introduction is given to the term “thermoelectric power” or “Seebeck coefﬁcient,” and some early research activities on DTEGs are summarized. The main part of this chapter deals with measurement techniques, modeling, and simulation of DTEGs based on semiconducting oxides. Recent results obtained with some semiconducting materials, together with a consideration of ionic DTEGs, complete the main part of this chapter. The chapter concludes with a discussion on the disadvantages and drawbacks of DTEGs and possible future research topics. 11.1.2 Thermoelectric power The intention of this section is to give the inexperienced reader a short introduction to the physical background of thermopower, also known as “thermoelectric power.” The reader is referred to Thermoelectricity25 for a more detailed analysis. The simplest treatment of the thermopower for semiconductors is based on the fact that the velocity of electrons increases with increasing temperature. Let us assume a wire that is divided along its length into a low-temperature section and a high-temperature section. In cross section, at the point in the middle where the two sections meet, all electrons that pass the interface are counted. After a certain time, more electrons will have traveled from the high-temperature section to the low-temperature section than vice versa. As a result, an electrical voltage evolves between both sections to compensate the driving force. In the book of Thermo- electricity25 a detailed calculation is given for a semiconductor assuming a Boltzmann distribution of the electrons and a temperature gradient in a certain direction. The thermovoltage is caused by the thermal diffusion of the charge carriers in the temperature gradient in the different sections of the wire. The calculation clearly demonstrates that temperature- dependent contact voltages do not play a role in the measured thermovoltage. As a result, the thermopower (Seebeck coefﬁcient), h, of pure n-type, pure p-type, and mixed nep-type conductors can be expressed, respectively, by Eqs. (11.1e11.3) (e.g., 26): kB NC h ¼ ln þ A (11.1) n conductor e n e 350 F. Rettig and R. Moos

kB NV h ¼ ln þ A (11.2) p conductor e p h

snh þ sph h ¼ n p (11.3) sn þ sp

In these equations, kB is the Boltzmann constant, e is the electron charge, NC and NV are the effective densities of states in the conduction band and in the valence band, Ae and Ah are the transport constants representing the scattering mechanism, n and p are the concentrations of electrons in the conduction band and holes in the valence band, and sn and sp are the conductivities of the electrons and the holes. The second, more general, treatment is based on nonequilibrium thermodynamics. Fluxes and forces are connected by a matrix. The diagonal elements (the main effects) of this matrix are well-knowndfor example, the diffusion coefficient (which is the connection between particle fluxes under a concentration gradient) or the thermal conductivity (which relates the temperature gradient with the heat flux). One of the nondiagonal elements is the Seebeck coefficient (thermopower, h), which relates a temperature gradient with a particle flux. Based on this, general equations are obtained that describe the heat and particle flow in a thermal and concentration profile:

divðs , gradVthermoelectric þ sh , gradTÞ¼0

divðshT , gradVthermoelectric þ k , gradTÞz0 (11.4)

Here, s is the conductivity, Vthermoelectric is the thermovoltage, h the Seebeck coefficient, T the temperature, and k is the thermal conductivity for a vanishing electrical field. The second equation is here set to zero, as Joule heating as a second-order effect does not play a significant role. More details can be found in Refs. 27 or 28. 11.1.3 Direct and indirect thermoelectric gas sensors Thermoelectric gas sensors can be divided more or less arbitrarily into direct and indirect thermoelectric gas sensors. Until recently, research has been mainly addressed to indirect gas sensors, and DTEGs have rarely been studied. Fig. 11.1 explains the working principle for both types of thermoelectric gas sensor. Indirect thermoelectric gas sensors use the heat of an exothermic reaction that stems from a combustible analyte. The temperature on a catalytically Semiconducting direct thermoelectric gas sensors 351

(a) Indirect thermoelectric gas sensor V C3H8 + 5O2

–ΔH

Catalytic ΔT 3CO + 4H O active coating 2 2 Thermoelectric material

(b) Direct thermoelectric gas sensor (absolute temperature measurement) V gsf Pt

Gas sensitive film T T 2 (gsf) 1

T T 2 Thermo couples (tc) 1 Au (c) Direct thermoelectric gas sensor (relative temperature measurement) V Pt gsf Pt

Gas sensitive film (gsf)

T Au T 2 1 Pt

V ΔT ΔT, Figure 11.1 Principle of the setup of (a) indirect and (b) and (c) direct thermoelectric gas sensors. The difference between (b) and (c) is the method of measuring the temperature difference DT. Reprinted from Rettig F. (2008), Direkte thermoelektische Gassensoren (in German) [PhD thesis]. University of Bayreuth, Shaker-Verlag; 2008 with permission from Shaker-Verlag. coated area of a (usually planar) substrate increases with the concentration of the analyte.29 The temperature difference between catalytically inactive areas on the substrate is usually measured either by thermocouples or by thermopiles. Therefore, this type of sensor is called an “indirect thermoelectric gas sensor.” Its measurement principle is similar to pellistor sensors.30 352 F. Rettig and R. Moos

The thermoelectric material itself should not be catalytically active. The catalytically active coating of the thermoelectric material defines both the sensitivity and the selectivity of the sensors. As an example, the reader is referred to Refs. 29 or 31, where further information on indirect thermoelectric gas sensors is given. In contrast, in DTEGs, the Seebeck coefficient (thermopower, h) of the gas sensitive material itself changes when the concentration of the analyte varies in the ambient atmosphere. The density of the free electrons and/ or defect electrons (holes)dor, in other words, the Fermi leveldis directly affected by a changing analyte gas concentration. There are several possible physical effects to explain how the Fermi level can be dependent on the gas phase. Chemisorption, for instance, may occur following the reaction: þ 4 O2 2e 2Oads (11.5) This chemisorption process captures electrons from the gas sensitive material. Therefore, a space charge region evolves from the interface of the material and the gas phase. In porous structures, this interface is typically the grain surface. For an n-type semiconductor such as SnO2, electrons are depleted in this region, resulting in an increased resistance not only for the whole grain but also for the whole gas sensitive film. If reducing gases are present, they may consume the chemisorbed oxygen and the electron is transferred back to the gas sensitive material, the space charge regions vanish partly, and the resistance of the gas sensitive material decreases, often by decades. Typically, at higher temperatures, one finds effects in which the bulk of the material is involved. The electron concentration in the bulk material can be modulated by ex- or incorporation of oxygen according to

1 ••4Ox O þ 2e þ V O (11.6) 2 2 o Then, the oxygen partial pressure of the surrounding gas atmosphere is the driver for a change in the electron concentration. These examples clearly show that DTEGs are based on the same physical principles as conductometric gas sensors, as in both cases the analyte concentration modulates the electron density. However, the measurand is different. In conductometric devices, the material property “conductivity” changes and, hence, the resistance of a sensor varies with the analyte concentration. In contrast, the determination of thermopower is more complicated, as not only the thermovoltage has to be measured but also a known temperature difference has to be applied or, at least, measured. Semiconducting direct thermoelectric gas sensors 353

11.1.4 Early research activities The concept of DTEGs is not a recent one. Several authors worked on this topic in the 1980s, but no systematic studies were conducted at that time. Pisarkiewicz and Stapinski32 reported on the change of the Seebeck coefficient of SnO2 when applying reducing gases. The effect was attributed to a modulation of the depletion layer at the grain surfaces affecting the Fermi 33 level. Siroky used a thermoelectric gas sensor based on SnO2 to detect flammable gases. Here, it was considered to measure thermopower and conductivity in parallel. Mizsei34 also explained the change of the thermoelectric power of palladium-activated tin oxide SnO2 in the presence of H2 with the affected depletion layer. Moos35 described a method for measuring the oxygen content of a gas by using the thermoelectric effect of a bulk 36 material. Ionescu reported on a SnO2 gas sensor with increased selectivity using simultaneous measurement of resistance and the Seebeck coefficient. 37 In 2000, Liess and Steffes presented a DTEG based on In2O3, and Smulko et al.38 used thermoelectric voltage fluctuations for gas sensing. However, it should be noted that this early work was scattered research, without a holistic consideration of the material and the correspondingly requisite transducers and evaluating systems. Additionally, these early approaches did not classify their devices as DTEGs.

11.2 Direct thermoelectric gas sensors The following sections present the optimization of the transducers and the gas sensitive materials as well as results for different DTEGs. Section 7.2.1 covers the measurement technique for DTEGs. Section 7.2.2 describes the theoretical design of transducers and gas sensitive materials to enable the design of accurate, fast, and long-term stable DTEGs. Section 11.2.3 presents results for different DTEGs with different materials. Section 11.2.4 describes ionic DTEGs as alternatives to semiconducting oxide materials. 11.2.1 Measurement techniques Compared with the relatively simple resistance measurement, DTEGs require a more sophisticated setup. The measurand thermopower (Seebeck coefﬁcient) is deﬁned by DV h ¼ h gsf (11.7) gsf Pt DT 354 F. Rettig and R. Moos

In Eq. (11.7), hgsf is the Seebeck coefficient of the gas sensitive film, DVgsf is the measured thermovoltage of the gas sensitive film, and DT is the temperature difference at the junctions between the gas sensitive layer and the conductor tracks. Owing to the fact that the conductor tracks also add a thermovoltage, the thermopower of the gas sensitive layer has to be corrected by the thermopower of the conductor track material (here platinum), hPt. There are two ways to determine the temperature difference between the junctions of the conductor tracks and the gas sensitive film. Fig. 11.1 depicts both possibilities: in (b), the temperature difference, DT, is directly determined, whereas in (c), the temperatures at both junctions, T1 and T2, are measured separately and the temperature difference DT is calculated. Because of the fact that not only the temperature difference but also the temperature of the gas sensor has to be controlled precisely, the option presented in Fig. 11.1(c) is advantageous. Combinations of both options are also possible. The thermovoltages of metallic thermocouples are usually easy to measure, although the voltages are in the microvolt range. In contrast, the internal resistances of semiconducting oxides are by orders higher. Such high ohmic voltage sources are difficult to measure.39,40 Therefore, a transducer for a DTEG has to be developed to ensure an accurate performance. The transducer which will be discussed below allows a maximum internal resistance of the gas sensitive layer of about 1 MU. According to Eq. (11.7), it would be possible to apply the temperature difference statically, but a temperature modulation technique enables one to measure the thermovoltages of the sensor material more precisely. In addition, some plausibility checks are possible. Furthermore, it is well- known that materials may decompose slowly in a temperature gradient due to the Soret effect.41 Fig. 11.2 shows the design of a thermoelectric gas sensor device manufactured according to planar ceramic multilayer technology. The heater brings the tip of the sensor to operation temperature by applying a heater voltage, Vheater. The modulation heater generates the temperature difference for the gas sensitive layer. For this purpose, a sinusoidal modulation voltage, Vmodu, is applied. The equipotential layer will be explained in the next paragraph. The gas sensitive layer and two thermocouples are located on the top of the sensor. With the help of the thermocouples, the temperatures Semiconducting direct thermoelectric gas sensors 355

Gas sensitive film and thermocouples V Insulation layer gsf T T Equipotential layer 1 2

Insulation layer

Modulation heater

Substrate V modu Heater

V heater

Thermocouple Thermocouple 4 mm Gas sensitive film Pt Equipotential ring Pt Figure 11.2 Setup of the direct thermoelectric gas sensors presented in this chapter. The applied and measured voltages are also indicated. Reprinted from Rettig F., Moos R. Temperature-modulated direct thermoelectric gas sensors: thermal modeling and results for fast hydrocarbon sensors. Meas Sci Technol 2009;20(6):0652059 with permission from IOP-Publishing Ltd.

T1 and T2 are determined. The thermovoltage of the gas sensitive ﬁlm, Vgsf, is measured over the platinum legs of the thermocouples. According to Eq. (11.7), the thermopower of the gas sensitive layer can be calculated. More information on the challenges faced in manufacturing such sensors can be found in Ref. 27. Fig. 11.3 shows experimental data relating to DTEGs: (a)e(d) stem from a sensor without an equipotential layer. The temperature difference in Fig. 11.3(a) is clearly sinusoidal, whereas the thermovoltage, Vgsf, of the gas sensitive layer differs signiﬁcantly from sinusoidal behavior. The distorted signal, Vgsf, prevents a linear regression with DT. The reason for the distortion becomes apparent from Fig. 11.3(d), where a Fourier analysis of the thermovoltage of the gas sensitive layer, Vgsf, is shown. Besides the expected 356 F. Rettig and R. Moos

(a) (b) (c) 0 4 4 6 2 2 –2 4 0 0 (K) T

–4 –2 –2 Δ 2 (mV) (K) –6 –4 –4 (d) 0 T gsf Δ –6 –6 8 –8 V 6

–8 –8 (mV) –10 4 –10 –10 gsf 2 –12 –12 –12 V 0 0 100 200 300 400 –12 –9–6 –3 0 5 10 15 2 0 t (s) ΔT (K) f (mHz) (e) (f) (g) 0 –0.5 –0.5 –1 4 –1.0 –1.0

–2 (K)

T 2 –3 –1.5 –1.5 Δ (K) (mV) 0 –4 –2.0 –2.0 (h) T Δ –5 gsf 2

–2.5 V –2.5

–6 (mV) –3.0 –3.0 1

–7 gsf –8 –3.5 –3.5 V 0 0 100 200 300 400 –8 –6 –4 –2 0 5 10 15 20 t (s) ΔT (K) f (mHz) Figure 11.3 Measured and analyzed signals of a direct thermoelectric gas sensor (a), (b), (c), (d) without equipotential layer and (e), (d), (f), (g), (h) with equipotential layer. The time resolved signals of the temperature difference, DT, and the measured thermovoltage, Vgsf, are shown in (a) and (e). The DT(Vgsf) diagram is given in (b) and (f). The result of the Fourier analysis of DT and Vgsf can be found in (c), (d), (g), and (h). signal of 10 mHz, an additional signal at 5 mHz is found. This is the frequency of the applied modulation voltage, Vmodu. According to Eq. (11.8), the temperature modulation frequency, fDT, is twice the modulation frequency, fmodu: 2 2 V V ; P ¼ modu ¼ 0 moud$cos2ðpf tÞ R R modu (11.8) 2 2 V ; V ; ¼ 0 modu$ð1 þ cosð4pf tÞÞ ¼ 0 modu$ð1 þ cosð2pf tÞÞ 2R modu 2R DT The 5 mHz signal is an interference of the modulation voltage, Vmodu, with the thermopower due to a small residual conductivity of the substrate material at elevated temperatures of several hundred degree Celsius. The finite resistance of the substrate material and the modulation voltage, in the range of several volts, in combination with the relatively high resistance of the gas sensitive layer and the small thermovoltages, indicates that the amplitudes of thermovoltage and the disturbing voltage are in the same Semiconducting direct thermoelectric gas sensors 357 order. The low resistance of the thermocouples is a short circuit for the disturbing modulation voltage and, therefore, the disturbing voltage collapses, in contrast to the situation where the gas sensitive layer has a significantly higher resistance. Therefore, the low resistance of the thermocouples compared with the high resistance of the gas sensitive layer is the reason why the temperatures measured by the thermocouples remain almost unaffected by the modulation. It would be possible to implement a Fourier analysis to calculate the thermopower of the gas sensitive layer from the distorted voltage of the gas sensitive layer,43 but an improved design of a DTEG with an additional equipotential layer offers the possibility to measure almost sinusoidal signals of the temperature difference, DT, and the thermovoltage of the gas sensitive layer, Vgsf. Fig. 11.3(e)e(h) shows the results of a sensor with an equipotential layer. Both the Fourier analysis of Vgsf and DT show contributions of nearly one frequency at 10 mHz (Fig. 11.3(g) and (h))d that is, the signals are almost sinusoidal (Fig. 11.3(e)) and, therefore, the slope of a linear regression can be used to determine the thermopower of the gas sensitive film (Fig. 11.3(f)). More details on the design of the equipotential layer can be found in Ref. 43. The higher the internal resistance of the gas sensitive material, the more necessary the equipotential layer becomes to improve the signal quality of the DTEGs. Although the modulation technique improves accuracy, there is a fundamental drawback of the temperature modulation (Fig. 11.3(e)): at least one complete modulation period is required to obtain a first reading for the thermopower of the material. With a frequency of 10 mHz, the response time of such a sensor is about 100 s. Even if the material itself responds much more rapidly to changing analyte concentrations, the measurement technique impedes a faster sensor response. The next section deals with a solution to the problem of reducing the response time and presents a simulation of the electrical and thermoelectric material properties. 11.2.2 Modeling and simulation of thermoelectric gas sensors According to Ref. 44, the thermal time constant (relaxation time after a sudden temperature step) of typical gas sensors manufactured in thick-film technology on planar alumina substrates is around 10 s. This thermal time constant is valid for large temperature steps. It is mainly driven by the convection coefficient of the gas sensor substrate. However, for temperature 358 F. Rettig and R. Moos modulation, large temperature differences are not necessary. A temperature difference of 20 Csuffices for a DTEG. To reduce the response time of the sensor, the temperature modulation frequency has to be increased significantly. The idea of a higher modulation frequency can be explained by a simple model known from Earth sciences. If one applies a sinusoidal temperature change to the ground, it is interesting to consider the depth to which ground temperature modulations are present. If one considers, for instance, a daily temperature change (hot days and cold nights), the penetration depth of the thermal wave is around 10 cm. For yearly temperature modulations (hot summers and cold winters), the thermal wave can penetrate to a depth of around 1 m. That is the reason why water pipes should be installed at least to this depth, otherwise the water would freeze. As a result, the frequency of the thermal modulation has a major influence on the penetration depth.45 Based on this concept, a one-dimensional model has been built up, which describes the thermal behavior of a DTEG. A more detailed explanation of the model can be found in Ref. 42. The setup of the one-dimensional model is shown in Fig. 11.4. Because of the fact that the ratio of the cross section to the perimeter is small, it is possible to introduce surface-related convective heat loss as a volume heat loss into the one-dimensional partial differential equation: v v2 $ r T k T ¼ 1 ð þ ð p ÞÞ hconv Psensor ð Þ cp 2 q0 1 exp 2j ft T Ta vt vx 2d Asensor for x < d = 2 (11.9) v v2 $ r T k T ¼hconv Psensor ð Þ cp 2 T Ta vt vx Asensor (11.10) for x d=2

Modulation heater Symmetry plane h sensor

b sensor

0 d/2 dx x Figure 11.4 Setup for the one-dimensional model for the thermal simulation of a direct thermoelectric gas sensor. Reprinted from Rettig F., Moos R. Temperature-modulated direct thermoelectric gas sensors: thermal modeling and results for fast hydrocarbon sensors. Meas Sci Technol 2009;20(6):0652059 with permission from IOP-Publishing Ltd. Semiconducting direct thermoelectric gas sensors 359

Eqs. (11.9) and (11.10) were used to simulate the thermal behavior of the sensor. The density of the material is r, cp is the heat capacity, T is the temperature, t the time, k is the thermal conductivity, x is the coordinate according to Fig. 11.4, d is the lateral length of the heater, q_ is the heat generation of the modulation heater per volume, f the modulation frequency, hconv the convection coefficient, Psensor the perimeter of the sensor (Psensor ¼ 2hsensor þ 2bsensor), Asensor the cross section of the sensor (Asensor ¼ hsensor $ bsensor), and Ta the ambient temperature. These partial differential equations can be divided into a static differential equation and into a differential equation for a stationary harmonic solution. Details can be found in Refs. 27 or 42. The solution to these equations is 8 > pffiffiffiffi Àpffiffiffiffi Á > d d > 1 exp K cosh K x for 0 < x < 2 2 ð Þ¼ G TM;S x 2K > pffiffiffiffi pffiffiffiffi > d À Á d :> sinh K exp K x for x > 2 2 q h P 2pf rc with G ¼ 0 and K ¼ conv sensor þ j p 2dk kAsensor k (11.11)

For the static part of the solution, TS, f is zero, otherwise the actual temperature modulation frequency, f, is used to calculate the amplitude and the phase angle of the complex temperature distribution, TM (x). The considered thermal model of the sensor was veriﬁed by comparison with the measured temperature distribution of a real sensor. The factor G relates the volume source of heat q_ with the geometry of the modulation heater. K is, in general, the decay constantddue to the fact that it is a complex number, a decaying thermal wave is the result. Fig. 11.5(b) shows the temperature difference amplitude, DTM (x), with the modulation frequency as a parameter. The experimental data were obtained by a line scan using an infrared camera. The mean applied electrical power was the same for all different modulation frequencies f. The static temperature distribution, Ts(x), therefore agrees for all temperature modulation frequencies (Fig. 11.5(c)). This is validated by the thermal model of the sensor and by the measured static temperature distribution. As expected, the amplitude of the harmonic thermal distribution decreases with increasing frequency (Fig. 11.5(b)). The length of the gas sensitive layer was designed to be 4 mm. If the modulation heater was placed at one end of 360 F. Rettig and R. Moos

(a)102 (b) 25 f = 0.31 Hz f 20 = 1 Hz f = 3.16 Hz h = 650 μm sensor f = 10 Hz 15 f 1 = 31.6 Hz 10 (°C)

M Model T

Δ 10

(mm) Standard parameters T

l 5

k 0 = 1.4 Wm 10 0 (c) 280

–1 260 K 240 –1

(°C) 220 s 200 T 180 10–1 160 10–3 10–2 10–1 100 101 01234567 f x mod (Hz) (mm) Figure 11.5 Results of the thermal modeling of a direct thermoelectric gas sensor. The decay constant (penetration depth) IT of the thermal is shown in (a). The good agreement of both the harmonic part and the static part between model and measurement is demonstrated in panels (b) and (c). (c) Reprinted from Rettig F., Moos R. Temperature- modulated direct thermoelectric gas sensors: thermal modeling and results for fast hydrocarbon sensors. Meas Sci Technol 2009;20(6):0652059 with permission from IOP-Publishing Ltd. the gas sensitive layer, the temperature at the other end of the gas sensitive layer would barely be affected by the harmonic part of the thermal modulation. The behavior is described best with the parameter “penetration depth” of a thermal wave lT. Fig. 11.5(a) shows lT as a function of the modulation frequency. A modulation frequency of f z 1 Hz results in a penetration depth of the thermal wave lT of around 1e2 mm. Therefore, a temperature modulation frequency of 1 Hz requires a modulation heater to be placed at a distance of less than 1 mm from one end of the gas sensitive layer. Otherwise, only a small amplitude of the thermal wave will reach the gas sensitive layer. A sensor was fabricated according to the design illustrated in Fig. 11.2. First, the platinum modulation heater was screen-printed and fired; then, the insulation layer was applied onto the modulation heater layer. After the equipotential layer and a further insulation layer, the heater and its conductor tracks were applied. Finally, the thermocouples were Semiconducting direct thermoelectric gas sensors 361 screen-printed on the top and fired, and an insulation layer was screen- printed and fired. As a result, the distance between the modulation heater and the thermocouple was around 40e60 mm, while the other thermocouple was 4 mm away. Experimental results with different DTEGs are discussed in the next section. The most important part of DTEGs is, of course, the gas sensitive material. Typical gas sensitive materials for classical conductometric gas sensors were not developed and optimized for the application as DTEGs. The Seebeck coefficients of these semiconducting oxide materials change with the concentration of free charge carriers, as shown in Eqs. (11.1)e(11.3).The measurand resistance (conductance) is always a positive valuedin contrast to the thermopower, which can have either positive or negative signs. This offers the opportunity to use different materials for DTEGs. Moos35 described a direct thermoelectric oxygen sensor with an intrinsic bulk material to be operated above 600 C. Another approach is considered here for simulation: gas sensitive materials for temperatures around 400e600 C. In this temperature range, bulk incorporation of oxygen in these materials is a very slow process and can be ignored.46 Simulation results obtained for semiconductor materials are shown in the next paragraphs. Fig. 11.6 depicts the geometrical model. The model has a two-dimensional rotational symmetry. The dark gray area is considered for the simulation. Each grain is described by its radius RK. The neck radius, RH, describes the interconnection with adjacent grains. Chemisorption takes place at the grain surface. A space charge region develops from the grain

R r R K H T Grad x 2ne– 2On–

O2 Chemisorption Figure 11.6 Model with rotational symmetry for the analysis of materials for direct thermoelectric gas sensors. The dark gray area was simulated with the commercial FEM-software Comsol Multiphysics. Reprinted from Rettig F., Moos R. Morphology dependence of thermopower and conductance in semiconducting oxides with space charge regions. Solid State Ionics 2008b;179(39):2299e2307 with permission from Elsevier. 362 F. Rettig and R. Moos surface to the middle of the grain. The overall thermoelectric and conductive properties of the grain were calculated as follows: First, the isothermal PoissoneBoltzmann equation was solved48 from Eq. (11.12): e2n p n N N divðgradFÞ¼ i 0 expð FÞ 0 expðFÞ A D ε0εrkBT ni ni ni (11.12) Here, F is the reduced potential, which is given by ef F ¼ (11.13) kBT where e is the elementary charge, 4 is the electrical potential, kB is the Boltzmann constant, and T is the temperature. In Eq. (11.12), ni is the intrinsic charge carrier concentration, ε0 $ εr is the dielectric constant, p0 and n0 are the hole concentration in the valence band and the electron concentration in the conduction band, respectively, and NA and ND are the acceptor and the donor concentration in the material. Eq. (11.12) is valid for materials in which electrons and holes are the mobile charge carriers. A precondition for the validity of the equation is that the conduction band and the valence band are only weakly occupied so that the Boltzmann statistic for the charge carriers can be applied. For the simulation, Eq. (11.12) was normalized regarding the space coordinates; details can be found in Ref. 47. The Debye length gives an idea of the extent of the space charge regions in the grain. For a p-type semiconductor, the Debye length LD,p can be calculated by sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ε0εrkBT ; ¼ LD p 2 (11.14) e p0 The normalized nonlinear partial differential equation (Eq. 11.12) was implemented for the rotational symmetric geometry according to Fig. 11.6 into the commercial finite element method program Comsol Multiphysics. The solution of this equation is the distribution of the reduced potential in a gas sensitive grain. An example solution is shown in Fig. 11.7. A reduced surface potential F of 5 was applied to the grain surface. At a temperature of 400 C, this reduced potential corresponds to an electrical potential 4 of around 300 mV. The grain radius was RK ¼ 200 nm, and the neck radius was Semiconducting direct thermoelectric gas sensors 363

r R = K

0 x Φ

–5 –4 –3 –2 –1 0 Figure 11.7 Obtained simulation data for the reduced potential F. A reduced surface potential was set to F ¼5 for a grain radius of RK ¼ 200 nm and a neck radius of RH ¼ 80 nm. The material was slightly donor-doped (ND/ni ¼ 10 and NA/ni ¼ 0.1). The Debye length was 20 nm. Reprinted from Rettig F., Moos R. Morphology dependence of thermopower and conductance in semiconducting oxides with space charge regions. Solid State Ionics 2008b;179(39):2299e2307 with permission from Elsevier.

RH ¼ 80 nm. The material itself was slightly donor-doped (ND/ni ¼ 10 and NA/ni ¼ 0.1) and the Debye length was 20 nm. The middle of the grain is barely inﬂuenced by the reduced surface potential, whereas almost the whole neck region is inﬂuenced by the reduced surface potential. Using Eqs. (11.15) and (11.16), the charge carrier concentration distributions in the grain can be calculated49:

n ¼ n0 expðFÞ (11.15)

p ¼ p0 expð FÞ (11.16) As a result, Fig. 11.8 shows the electron concentration (top) and the hole concentration (bottom) in the different areas of the grain. At each point of $ ¼ 2 the grain, n p ni is valid. However, as the electron concentration can never be lower than zero, the space charge region of the electrons extends much more toward the grain center than the space charge region of the holes. The reduced potential F, the electron concentration n/ni, and the hole concentration p/ni are extracted on the r-axes from Fig. 11.8 and plotted in Fig. 11.9. Starting from the grain surface, an inversion area with a length of about 30 nm can be seen. In this area, the hole concentration is larger than the electron concentration, although the material is slightly donor-doped. This inversion plays a major role for the enhanced sensitivity of materials for DTEGs. 364 F. Rettig and R. Moos

r R n n = K 0 n n i 14 i p 12 n i n 10 0 = 10 n i x 8 p r R = K n 6 i

p 2 0 = 0.1 n n i i 0 x

Figure 11.8 Calculated electron concentration n/n0 and hole concentration p/p0 for a grain radius of RK ¼ 200 nm and a neck radius of RH ¼ 80 nm based on the results of Fig. 11.7. The material is slightly donor-doped (ND/ni ¼ 10 and NA/ni ¼ 0.1). The Debye length is 20 nm. Reprinted from Rettig F., Moos R. Morphology dependence of thermopower and conductance in semiconducting oxides with space charge regions. Solid State Ionics 2008b;179(39):2299e2307 with permission from Elsevier.

0 15 Φ

10 i Inversion area Inversion

–2 n / p

n n , Φ / i i n / n 5 p n –4 / i

0 0 50 100 150 200 r (nm)

Figure 11.9 Calculated course of the electron concentration n/ni and the hole concentration p/ni for a grain radius of RK ¼ 200 nm and a neck radius of RH ¼ 80 nm. The curves are extracted from Fig. 11.8 on the r-axis. The material is slightly donor-doped

(ND/ni ¼ 10 and NA/ni ¼ 0.1). The Debye length is 20 nm. Reprinted from Rettig F. Dir- ekte thermoelektische Gassensoren (in German) [PhD thesis]. University of Bayreuth, Shaker-Verlag; 2008 with permission from Shaker-Verlag. Semiconducting direct thermoelectric gas sensors 365

Based on Fig. 11.8, the local Seebeck coefﬁcients and the local conductivities can be calculated by applying Eqs. (11.1)e(11.3) and (11.17)e(11.19). s ¼ m n e nn (11.17) s ¼ m p e pp (11.18)

s ¼ sn þ sp (11.19) The calculation of the local properties is possible for homogeneous semiconductors; for inhomogeneous semiconductors, the presumptions have to be checked. The space charge region extends to about 20 nm. Therefore, the local properties change significantly in this length scale. The mean free path of the charge carriers has to be significantly lower than the width of the space charge region, as otherwise the charge carriers are not able to achieve the local equilibrium when traveling through the crystal. This assumption is definitively not valid for so-called “lifetime” semiconductors,50 where the restoration of the electronehole equilibrium takes a considerable time. However, in small polaron conductors, the electrons are more or less localized to single atoms and, hence, the assumption for the localized properties can be valid. For some oxide materials, a polaron-type conduction mechanism can be considered. Data for charge carrier lifetime or charge carrier mean free paths in oxides are rare. One of the few data available is published by Barsan;51 the mean free path of the electrons in SnO2 is lower than the Debye length by at least a factor of 25. ðs $ 4 Þ¼ div grad i 0 (11.20)

divðs $ gradVni þ sh $ gradTÞ¼0 (11.21)

divðshT $ gradVni þ k $ gradTÞ¼0 (11.22) Using the local properties and Eqs. (11.20)e(11.22) (from Ref. 28), the effective (overall) conductivity and the effective (overall) thermopower are calculated. The starting point for the calculation was the reduced potential F applied on the grain surface. This surface potential is the result of the chemisorption of oxygen on the grain surface. By integration of the space charge region, the surface charge concentration can be calculated. This surface charge concentration is a (nonlinear) function of the chemisorbed oxygen and, therefore, a measure for the concentration of oxygen in the ambient atmosphere. The chemisorption itself can be describeddfor 366 F. Rettig and R. Moos example, by a Wolkenstein isotherm.52 However, the correlation between the adsorbed amount of oxygen and the ambient oxygen concentration will not be covered here. It is sufﬁcient to take a look at the surface charge concentration to extract some interesting results regarding new materials for DTEGs. Fig. 11.10 shows the calculated effective thermopower, heff, as a function of the normalized surface charge concentration, which is (as explained in the last paragraph) a measure for the ambient oxygen concentration. It was calculated by dividing the thermovoltage difference from the left and the right grain boundaries by the temperature difference at the left and the right grain boundary. The normalization factor is the intrinsic charge carrier density, ni. Each element of Fig. 11.10 shows the thermopower for a differently doped material with the grain size as a parameter. The slope of the curve is a measure for the sensitivity of the material. A steep slope indicates

Intrinsic Slightly donor-doped Donor-doped 1.2 1.0 1.0 –0.9 )

–1 0.5 0.8 –1.0 0.6 0.0 (mVK –1.1

eff. 0.4 –0.5 η –1.2 0.2 –1.0 0.0 0.0 0.1 0.2 0.3 0.0 0.2 0.4 0.6 0153045 N ·n –1 μ N ·n –1 μ N ·n –1 μ Q i ( m) Q i ( m) Q i ( m) Slightly acceptor-doped Acceptor-doped 1.30 0.90 1.25 0.85 ) 1.20

–1 0.80 50 nm 1.15 0.75 1.10 200 nm (mVK 0.70 1.05 1000 nm eff. 0.65 η 1.00 0.60 0.95 0.55 0123 0 100 200 300 N ·n –1 μ N ·n –1 μ Q i ( m) Q i ( m)

Figure 11.10 The effective thermopower, heff, as a function of the surface charge 1 concentration NQ$ni . In each panel, the doping concentration is varied (intrinsic: ND/ni ¼ 0.1, NA/ni ¼ 0.1; slightly donor-doped: ND/ni ¼ 10, NA/ni ¼ 0.1; donor-doped: ND/ni ¼ 1000, NA/ni ¼ 0.1; slightly acceptor-doped: ND/ni ¼ 0.1; NA/ni ¼ 10; acceptor- doped: ND/ni ¼ 0.1, NA/ni ¼ 1000). The reduced surface potential varies from 0 to 5. Each curve is plotted for different grain radii. Reprinted from Rettig F. Direkte thermoelektische Gassensoren (in German) [PhD thesis]. University of Bayreuth, Shaker-Verlag; 2008 with permission from Shaker-Verlag. Semiconducting direct thermoelectric gas sensors 367 a large sensitivity. As illustrated in Fig. 11.10, acceptor-doped materials exhibit a low sensitivity, donor-doped materials show a better sensitivity, while intrinsic or slightly donor-doped materials are maximally sensitive. In addition, the grain size also influences the sensitivity. The grain radius of RK z 50 nm is near the Debye length and, therefore, the space charge region influences almost the whole grain. For large grains (RK z 1000 nm), only the surface regions are affected. Then, the effective thermopower is insensitive to changes of the ambient oxygen concentration. The slightly donor-doped material with a grain radius of RK z 50 nm has the maximum sensitivity. The reason is obvious from Fig. 11.9: if oxygen is adsorbed, an inversion layer is built from the surface of the grain. If a semiconductor changes from n-type to p-type semiconducting behavior, the sign change in thermopower (the sensitivity) reaches its maximal value. However, if one considers the measurand “resistance change” of a slightly donor-doped or intrinsic material, one finds almost no sensitivity, as the conductivity always has a positive sign. As a conclusion for this section, the semiconducting materials for DTEGs should be intrinsic or slightly donor-doped for a maximal sensitivity, if they are based on the chemisorption of oxygen. For bulk materials, where the entire stoichiometry of the material is changed, intrinsic materials may have also some advantages regarding sensitivity because, in this case, the sign of the thermopower also changes.26,53 However, many of the oxide materials that are typically used for gas sensors show a significant ionic conductivity at and around the intrinsic minimum.54 It is supposed that the ionic contribution will interact with the thermopower of electrons and holes. 11.2.3 Measurements and results This section deals with real transducers and gas sensitive materials for DTEGs. Using the results from previous sections, accurate, rapid, and long-term stable DTEGs with increased sensitivity can be designed. A first experiment to demonstrate the advantages of DTEGs is shown in Fig. 11.11.55 Here, instead of a planar sensor, a small porous ceramic brick-shaped sample of SrTi0.6Fe0.4O3ed was measured, as described below. The thermopower, h, and the resistance, R, were measured simultaneously (details in Ref. 56). The samples were kept at 700, 800, and 900 C for 7 h each. Within these 7 h (duration of each run), the oxygen partial pressure was varied stepwise, and the final values of R and h were plotted. It is interesting to observe that the resistance characteristics of the material shifted from run to run, presumably because of the sintering process of the sample. 368 F. Rettig and R. Moos

(a)2.2 (b) 20 0 First run First run Second run Second run 2.1 Third run Third run 180 2.0

1.9 ) –1 160 ) Ω VK / μ (

R 1.8 log(

STF40 140 1.7 η

1.6 T = 800°C Material STF: 120 SrTi Fe O δ 1.5 0.6 0.4 3–

1.4 100 –3 –2 –1 0 –3 –2 –1 0 p p log O2 (bar) log O2 (bar) Figure 11.11 (a) Resistance R and (b) thermopower (Seebeck coefﬁcient), h, of a porous

STF (SrTi0.6Fe0.4O3ed) specimen when exposed to different oxygen partial pressures. From Moos R, Izu N, Rettig F, Reiß S, Shin W, Matsubara I. Resistive oxygen gas sensors for harsh environments. Sensors 2011;11(4):3439e3465.

This large shift amounts to an error of approximately one decade in pO2. The thermopower, h, however, remains constant. This experiment clearly highlights the advantage of the DTEGs, which is based on the geometry independence of the potential difference measurement. As a first planar approach, a DTEG for hydrocarbons based on SnO2 is presented. In this case, the entire sensor was still heated in a tube furnace and only the temperature modulation was applied by a planar structure. Therefore, the manufacturing procedure of this sensor was simpler compared with the sensor presented in Fig. 11.2. Fig. 11.12(a) shows the design of the sensor. The upper part of the sensor containing the thermocouples and the gas sensitive layer was joined with the lower part with the modulation heater by the wet screen-printed equipotential layer. After drying and firing, the sensor was complete. The sensor was first tested with propane and then a significant part of the gas sensitive layer was milled out. The milled-out portion of the gas sensitive layer can be seen in Fig. 11.12(b). After milling out the gas sensitive layer, the sensing response was measured again. Fig. 11.13 shows the results. The thermopower of the gas sensitive layer, hSnO2, is barely influenced after milling out a portion of the gas sensitive Semiconducting direct thermoelectric gas sensors 369

(a) Gas sensitive layer and reference

Alumina substrate

Equipotential layer Pt

Alumina substrate

Modulation heater

(b) SnO2 layer Pt-conductor tracks Milled out part

Reference Equipotential ring

Figure 11.12 SnO2-based direct thermoelectric gas sensor (a) in thick-ﬁlm technology. (b) The sensor was measured before and after milling out a part of the gas sensitive layer. Reprinted from Rettig F, Moos R. Direct thermoelectric hydrocarbon gas sensors based on SnO2. IEEE Sens J 2007b;7:1490e1496 with permission from IEEE. © 2007 IEEE.

layer (Fig. 11.13(a)), but the resistance RSnO2 isdas expected from the preceding discussiondsigniﬁcantly increased (Fig. 11.13(b)). If a propane concentration of 100 ppm were present in the ambience of the gas sensor, a milled-out DTEG would measure a concentration of 80 ppm propane. However, a milled-out conductometric gas sensor (Fig. 11.13(b)) would only measure a concentration of 30 ppm propane. One sees also a drift of both measurands: thermopower and resistance. At ﬁrst glance, they correlate. Such an assumption can be checked with a 370 F. Rettig and R. Moos

(a) (b) –380 106

–400 Ω) ( 2 SnO

–420 R 105 ) –1 –440 0 20406080100 VK

μ c

( (ppm) C3H8 2 (c) –460

SnO –380

) –400 –480 –1 –420 VK

μ –440 (

2 –460 –500 Before milling out –480 SnO After milling out η –500 –520 –520 0 20406080100 10–6 10–5 10–4 c (ppm) –1 C3H8 R (Ω–1) SnO2 Figure 11.13 Measurement results of a direct thermoelectric gas sensor as shown in Fig. 11.12, with SnO2 as the gas sensitive material at 400 C with 1% oxygen with different propane concentrations (balance nitrogen). The curves indicate (a) the results of the thermopower hSnO2, (b) the resistance, RSnO2, and (c) the Jonker diagram, before (circles) and after (triangles) milling out a part of the gas sensitive layer. Reprinted from Rettig F, Moos R. Direct thermoelectric hydrocarbon gas sensors based on SnO2. IEEE Sens J 2007b;7:1490e1496 with permission from IEEE. © 2007 IEEE.

58 h Jonker diagram. In such a plot, the thermopower, SnO2, and the conductivity (or the resistance, RSnO2) are plotted against each other. If, after milling out areas of the ﬁlm, the data points stay on the same curve as the points before milling out, the assumption would be correct. However, this is not the case, as the curves in Fig. 11.13(c) are clearly shifted to the right after the milling process. The measurand thermopower is not a function of the geometry of the gas sensitive layer. This might be advantageous for abrasive gas streams. More details of this sensor can be found in Ref. 57. Fig. 11.14 shows photographs of two DTEGs with SnO2 as the gas sensitive ﬁlms. The sensors were manufactured as shown in Fig. 11.2. The gas sensitive layer was applied with a brush to ensure low internal resistance of the gas sensitive layer (because of the geometry independency of the measurand thermopower, the geometry does not play a role!). The sensors also had a heater on the reverse side. It heated the entire sensor tip to the operational temperature of 400 C. The temperature modulation was Semiconducting direct thermoelectric gas sensors 371

SnO2

Figure 11.14 Photograph of direct thermoelectric gas sensors with SnO2 as the gas sensitive layer. Reprinted from Rettig F, Moos R. Temperature-modulated direct thermoelectric gas sensors: thermal modeling and results for fast hydrocarbon sensors. Meas Sci Technol 2009;20(6):0652059 with permission from IOP-Publishing Ltd. applied by the modulation heater with a modulation frequency of 0.312 Hz. A continuous regression was used to extract the thermopower of the gas sensitive layer. Fig. 11.15 shows the results for both sensors. The propane concentration proﬁle is shown in Fig. 11.15(a), and the measured thermopower of the gas sensitive layer is plotted in Fig. 11.15(c). Fig. 11.15(b) shows the Dh h corresponding error of the thermopower, SnO2 SnO2, determined by the error of the regression analysis. The characteristics of the sensors can be found in Fig. 11.15(d). Both sensors behave nearly identically, in the transient diagram and in the sensor characteristics. For a resistive gas sensor, identical behavior would be surprising, especially when the gas sensitive layer was applied using such a poorly reproducible technique. The errors Dh h SnO2 SnO2 from the continuous regression are usually below 1%. When considering the modulation frequency of 0.312 Hz, this low error is remarkable. The error can be used to check if the sensor is working correctly. If the error is above a certain level for a certain time, the sensor has to be checked. This clearly shows that accurate, rapid, and reliable DTEGs can be designed. More details about this sensor can be found in Ref. 42. As shown above, bulk materials also show a promising oxygen gase dependent thermoelectric behavior. Therefore, two DTEGs were prepared 372 F. Rettig and R. Moos

(a) 600 500 400

(ppm) 300 200 H 100 C

c 0 (b) 1.6 1.2 (%) 0.8 0.4 SnO SnO

η η Δ 0.0 (d) (c) – 350 350

– 400 400

) – 450 450 )

VK – 500 VK μ 500 μ ( (

– 550 550 SnO SnO

η η – 600 First sensor 600 Second sensor – 650 650

0 500 1000 1500 0 150 300 450 t ( C s) C3H8 (ppm)

Figure 11.15 Results obtained from a direct thermoelectric gas sensor with SnO2 (see Fig. 11.14) at 400 C and 1% oxygen with different propane concentrations (balance nitrogen). The temperature modulation frequency was 0.312 Hz. The curves indicate (c) the results of the thermopower hSnO2, (b) the relative error of the thermopower Dh h , and (d) the characteristics of the two gas sensors shown in Fig. 11.13. SnO2 SnO2 The transient propane profile is shown in (a). Reprinted from Rettig F, Moos R. Temperature-modulated direct thermoelectric gas sensors: thermal modeling and results for fast hydrocarbon sensors. Meas Sci Technol 2009;20(6):0652059 with permission from IOP-Publishing Ltd. with inks made from the same ceramic powder with which the brick-like specimens of Fig. 11.11 were made. Fig. 11.16(b) shows a photograph of two sensors with SrTi0.6Fe0.4O3ed as the oxygen gas sensitive material. The design of the sensor (Fig. 11.16(a)) was modified for this material because the diffusion barrier material SrAl2O4 is needed to prevent an 18 interaction of the gas sensitive material SrTi0.6Fe0.4O3ed with the alumina substrate during firing. The SrAl2O4 has to be fired at 1300 C for good adhesion. The printed insulation layer introduced in Fig. 11.2 is not suitable for such high firing temperatures. For the upper part of the sensor, the SrAl2O4 was printed and fired first on the alumina substrate. Then, the Pt-conductor tracks were printed and fired. Afterward, SrTi0.6Fe0.4O3ed Semiconducting direct thermoelectric gas sensors 373

(a) Pt-conductor tracks

SrTi0.6Fe0.4O3 layer Au-conductor tracks

SrAl2O4 layer

Substrate Al2O3 Equipotential layer Au Insulation layer

Modulation heater Pt

Substrate Al2O3

Heater Pt Heater conductor tracks Au

(b) Pt

SrTi0.6Fe0.4O3

Figure 11.16 Direct thermoelectric gas sensors with SrTi0.6Fe0.4O3ed as the gas sensitive material: (a) design and (b) photograph. Reprinted from Rettig F. Direkte thermoelektische Gassensoren (in German) [PhD thesis]. University of Bayreuth, Shaker- Verlag; 2008 with permission from Shaker-Verlag. paste was either screen-printed or applied with a brush. Finally, for the upper substrate, the Au-conductor tracks and the equipotential layer were screen-printed and fired in a separate step. The lower alumina substrate was printed with the heater, followed by the modulation heater. The heater conductor tracks and one insulation layer were screen-printed and fired in a single step. Both substrates of the sensor were joined by applying a second wet screen-printed insulation layer on the upper substrate. After drying and firing, the sensor was ready for the measurements. Details on the complex manufacturing process of this sensor can be found in Ref. 27. Please note the different methods by which the gas sensitive layer was applied: with a brush (Fig. 11.16(b), upper sensor, first sensor) and by screen printing (Fig. 11.16(b), lower sensor, second sensor). Both sensors were heated to their operational temperature of 700 Cby the Pt heater. A temperature modulation frequency of 0.156 Hz was applied 374 F. Rettig and R. Moos to the modulation heater. The thermopower was determined by a continuous regression analysis over two periods. Owing to this low modulation frequency, the sensor response time is limited to 12.8 s. The sensors were tested with different oxygen/nitrogen mixtures. Fig. 11.17 shows the results of both sensors as described in Fig. 11.16. The oxygen partial pressure was varied stepwise from pure nitrogen to pure oxygen (Fig. 11.17(a)). Both sensors behaved almost identically despite the fact that the geometry of the gas sensitive layer differed significantly (Fig. 11.16(b)). Both sensors reached their equilibrium state at each oxygen

(a)

101 100 10–1 (bar) –2

O 10 p 10–3 10–4 (b) 8 6 (%) 4 STF STF

η η Δ 2 0 (d) (c) 220 220 First sensor

200 Second sensor 200 180 180 ) ) VK VK 160 160

μ μ μ

( VK ( dec STF 140 140 STF η η

120 120

100 100

80 80 80 400 800 1200 10–3 10–2 10–1 100 t (s) pO (bar) Figure 11.17 Thermopower results of a direct thermoelectric gas sensor with a SrTi0.6Fe0.4O3ed gas sensitive film at 700 C in different oxygen concentrations (balance nitrogen). The temperature modulation frequency was 0.156 Hz. The black and the gray curves indicate (c) the results of the thermopower, hSTF, (b) the relative error of the thermopower, DhSTF/hSTF, (d) the characteristics of the two gas sensors shown in Fig. 11.16. The transient oxygen profile is shown in (a). Reprinted from Rettig F. Direkte thermoelektische Gassensoren (in German) [PhD thesis]. University of Bayreuth, Shaker-Verlag; 2008 with permission from Shaker-Verlag. Semiconducting direct thermoelectric gas sensors 375 concentration within 50 s. The relative error of the measured thermopower (Fig. 11.17(b)) was usually below 2%. However, the gas sensitive material had a relative low sensitivity toward oxygen. The slope in a half- 1 logarithmic plot is about 28 mVK per decade pO2. The material is known to have a slope of about 0.2 in a double logarithmic plot of the resistance versus the oxygen partial pressure. Therefore, the expected slope of a DTEG would be about 40 mVK 1 per decade.59 As already mentioned, the hole concentration changes in different oxygen concentrations. As a result, both the thermopower and the conductivity change. For a p-type material, like the one shown here, the slope in a logelog plot of resistance versus oxygen partial pressure can be transferred to a slope in a plot of the thermopower and the logarithmic oxygen partial pressure. In Fig. 11.11, this value was almost found. The reason for this deviation from the theoretically expected slope is not clear. However, if one compares Figs. 11.11 and 11.17 more closely, one finds that also the absolute value of the thermopower in Fig. 11.17 is lower than that in Fig. 11.11, e.g., h (1 bar, Fig. 11.11) z 108 mVK 1, whereas h (1 bar, Fig. 11.17) z 85 mVK 1. This behavior was also found in Ref. 60. In this case, it was assumed that the temperature difference was not measured at the same point where the thermovoltage was determined. In other words, the temperature difference (DT in Eq. 11.4) was determined too large; hence, the thermovoltage, as well as the slope in the h (pO2) plot, appeared too small. Therefore, in Ref. 60, a geometry correction factor was introduced and proven. Assuming the geometrical situation and a linear temperature gradient, the slope and the absolute thermopower could be increased by a factor of 1.1.27 However, the values of the bulk sample are not achieved. The different sintering temperature for the screen-printed layer (1100 C) and the bulk sample (1300 C) might be a reason. Also, the ionic thermopower of the material54 may contribute to the deviation from theory because, despite the material being a p-type semiconductor, ionic conductivity contributes to the electrical charge transport in a nonnegligible way. Overall, SrTi0.6Fe0.4O3ed can be used for DTEGs; however, the sensitivity is low. The last sensor presented in this section has much higher sensitivity. According to the previous section, intrinsic gas sensitive thermoelectric materials exhibit a larger sensitivity. As intrinsic material, Fe2O3 was used, as this material is known to have intrinsic semiconducting properties at low temperatures.61,62 This intrinsic semiconducting behavior makes the material unsuitable for conductometric gas sensors, as there is almost no change in conductivity with changing ambient pO2. However, it is an ideal 376 F. Rettig and R. Moos

(a) (b) Gas sensitive layer and thermocouples Thermocouple 1 Insulation layer

Equipotential layer Gas sensitive layer Insulation layer

Modulation heater Thermocouple 2 Alumina substrate Main heater Cover layer (c) 50 mm 6 mm

Figure 11.18 Direct thermoelectric gas sensors with Fe2O3 as the gas sensitive material: (a) design and (b) photograph. Reprinted from Rettig F, Moos R. a-iron oxide: an intrinsically semiconducting oxide material for direct thermoelectric oxygen sensors. Sensor Actuator B Chem 2010;145(2):685e690 with permission from Elsevier. candidate for DTEGs. The design of the sensor (Fig. 11.18(a)) is quite similar to that shown in Fig. 11.2. The conductivity of an intrinsic semiconducting material is typically low because only the intrinsic charge carriers contribute to the electrical conduction. The developed transducer for a DTEG allows a maximum internal resistance of the gas sensitive material of about 1 MU.If this range is exceeded, the measured thermovoltage becomes too noisy because of disturbance from ambient voltages. For Fe2O3, the internal resistance of the gas sensitive layer is higher than 1 MU with a gas sensitive layer 4 mm long. For this reason, the distance between the two thermocouples was reduced to about 1.2 mm. As a result, the internal resistance is sufﬁciently low for an accurate evaluation of the thermopower. Unfortunately, however, the reproducibility of different DTEGS suffers because of the small distance between the thermocouples. The point where the temperature difference is determined is typically not the same point at which the thermovoltage of the gas sensitive layer is read out. The gas sensitive layer was applied to the transducer with a brush (Fig. 11.18(b)). Fig. 11.18(c) shows a photograph of the complete gas sensor. The sensor was heated up to 580 C and a temperature modulation frequency of 0.312 Hz was applied. The continuous linear regression to Semiconducting direct thermoelectric gas sensors 377

(a) 10 10 10 (bar) 10 O

p 10 10 5 (b) ) 4 VK

μ 3 | (

O 2

Fe 1 η |Δ 0 (d) (c) 0 0

–100 dec –100 )

μVK ) 85 –200 –200 VK VK μ μ ( ( O –300 –300 O Fe First sensor Fe η η –400 Second sensor –400 First sensor Third sensor –500 –500 150 300 450 600 10 10 10 10 t (s) pO (bar)

Figure 11.19 Experimental results of the direct thermoelectric gas sensor with Fe2O3 as an oxygen sensitive layer at 580 C (sensor from Fig. 11.18). The temperature modulation frequency was 0.312 Hz. Fig. 11.19(a) shows the transient oxygen proﬁle,

(b) shows the absolute error of the thermopower, DhFe2O3, (c) depicts the transient result of the thermopower, hFe2O3, and (d) gives the characteristics of the three gas sensors. Reprinted from Rettig F, Moos R. a-iron oxide: an intrinsically semiconducting oxide material for direct thermoelectric oxygen sensors. Sensor Actuator B Chem 2010; 145(2):685e690 with permission from Elsevier. determine the thermopower was carried out for two periods; therefore, the response time of the sensor is limited to a regression time of 6.4 s. The temperature difference on the gas sensitive layer varied from 15 to þ5 C with respect to the sensor temperature of 580 C. Within this temperature difference range, the thermovoltage, DVgsf, correlated almost linearly with the temperature difference, DT. The temperature at both ends of the gas sensitive layer and the thermovoltage of the gas sensitive layer were determined with a 90 ms interval. Fig. 11.19 shows the results of a typical measurement run. The oxygen concentration was varied from pure nitrogen to pure ox- h ygen in seven steps (Fig. 11.19(a)). The determined thermopower, Fe2O3, varied from about 400 mVK 1 to 50 mVK 1 (Fig. 11.19(c)). The sensor reaches its equilibrium state within a few seconds. A more detailed timely analysis of the sensor response concluded that the measured response time 378 F. Rettig and R. Moos

(t63%) of 16 s can be assigned to the gas exchange in the test chamber. The Dh absolute error of the thermopower Fe2O3 (Fig. 11.19(b)) is usually below 2 mVK 1. The error exceeds this upper limit only after the stepwise of the oxygen concentration changes. The sensor is quite rapid, even the irregular- ities in the gas dosing at t ¼ 400 s are partially determined by the gas sensor. The major advantage of this gas sensor is its far higher sensitivity compared with the sensor shown in Fig. 11.17 (compare Fig. 11.19(d) with Fig. 11.17(d)). The sensitivity reaches a value of about 85 mVK 1 per decade and is about three times higher compared with the sensor based on SrTi0.6Fe0.4O3ed. More details on this sensor can be found in Ref. 63. It was shown in this section that it is possible to manufacture accurate, rapid, and sensitive DTEGs. The design of the DTEGs can be developed knowledge-based. As intrinsic materials show the best sensitivity, the internal resistance of the gas sensitive layers has to be considered, and the insulation and equipotential layers have to be applied. An appropriate temperature modulation frequency needs to be selected to achieve good results. 11.2.4 Ionic direct thermoelectric gas sensors The DTEGs introduced in the preceding sections were based on semiconducting oxide materials. However, other materials besides electronic conductors can be employed as materials for DTEGs. Several years ago, it was shown that the thermopower h of an electrochemical cell with Pt electrodes separated by an oxygen ion conductor follows Eq. (11.23) (e.g., 64): • k Q 2 h ¼ SðTÞ B lnðpO Þ O h (11.23) 4e 2 2eT Pt • In Eq. (11.23), QO2 is the heat of transport of the oxygen ions, S(T) is the entropy term, and hPt is the Seebeck coefﬁcient of platinum. In rough ﬁrst- order approximation, all these three terms can be considered as constant with respect to pO2. Then, a theoretical sensitivity s of the thermoelectric cell can be derived: dh k s ¼ ¼ B ln10 (11.24) d logðpO2Þ 4e According to Eq. (11.24), the sensitivity should be s z 50 mVK 1 per decade pO2, which is in the same order of magnitude as that of semiconductor materials. Semiconducting direct thermoelectric gas sensors 379

600 700°C 750°C 800°C 550 ) –1

(µVK 500 YSZ

–1 450 –50 μVK per decade

–3.0 –2.5 –2.0 –1.5 –1.0 –0.5 0.0 0.5 p log O2 (bar) Figure 11.20 Results of an ionic direct thermoelectric gas with yttria-stabilized zirconia as an oxygen sensitive layer. Note the almost nonexistent temperature dependence of the sensor signal. Reprinted from Roder-Roith€ U, Rettig R, Roder€ T, Janek J, Moos R, Sahner K. Thick-ﬁlm solid electrolyte oxygen sensors using the direct ionic thermoelectric effect. Sensor Actuator B Chem 2009;136(2):530e535 with permission from Elsevier.

The ﬁrst implementation of such a sensor device is reported in Ref. 60, in which 8 mol% Y2O3-stabilized zirconia was used for the gas sensitive material. The sensor setup was similar to the sensors described above; however, an additional Pt-cermet was applied to get a high exchange rate at the YSZePt interface. It becomes clear from the results in Fig. 11.20 that the sensitivity reaches the expected value. Astonishingly, but in accordance with Eq. (11.24), almost no temperature dependency of h was observed. This indicates that the three pO2-independent terms either have a negligible temperature dependency or their temperature dependencies compensate each other. Additionally, no cross-sensitivities to NO, H2, H2O, CO, CO2, or HC are observed. However, the long-term stability of this sensor has to be improved. The perovskite-type proton conductor BaCe0.95Y0.05O3ed has also been considered for ionic direct thermoelectric gas sensors.65 Although a hydrogen-dependent thermopower could be measured, the different mobile species (ions, electrons, holes) allow the material to apply only in certain atmospheres with deﬁned oxygen and hydrogen partial pressures. 380 F. Rettig and R. Moos

11.3 Conclusion and future trends DTEGs are an alternative to resistive gas sensors. Accurate, rapid, and long-term stable gas sensors have been presented in this chapter. The main advantage of DTEGs is the measurand “thermopower” or “Seebeck coefficient.” In contrast to conductometric gas sensors, the measurand thermopower is not influenced by changes in the geometry of the gas sensitive layer. A damage of the gas sensitive layer directly influences the resistance, but the thermopower remains virtually unaffected. An example of such an abrasion-resisting gas sensor is shown in Fig. 11.20. Besides the sensors discussed, DTEGs have been developed with special respect to the measurement principle. First, an adequate transducer has been developed. For the DTEGs presented, a temperature modulation technique was chosen to determine the thermopower. The advantage of this technique is improved accuracy and that the signals of the temperature differences and the thermovoltages can be analyzed either by regression analysis or by a Fourier analysis. Disturbing voltages are filtered out by these signal analyses. The disadvantage of the temperature modulation is the long response time of the sensor, which is determined here by the regression or the Fourier analysis. The problem can be overcome by rapid temperature modulation with a modulation heater placed within a distance of about 60 mm from one end of the gas sensitive layer. The thermal behavior of a DTEG has been modeled; both the thermal model and measurement of the thermal properties of the DTEGs agree very well. The gas sensitive layer of a DTEG determines the performance of the gas sensor. A general analysis of materials with chemisorption has been introduced. The model is based on semiconductor and thermoelectric equations. The partial differential equations (here PoissoneBoltzmann equations) have been solved by Comsol Multiphysics. The solution has been used to calculate the isothermal and the nonisothermal properties of a gas sensitive grain. The results of the simulation concluded that small grains are generally advantageous because of higher sensitivity. Furthermore, materials with only an intrinsic charge carrier density should have the largest sensitivity compared with n-type or p-type semiconducting materials. Four different gas sensors based on SnO2, SrTi0.6Fe0.4O3ed,Fe2O3, and YSZ demonstrated the potential of DTEGs. The classical material SnO2 was tested with propane, while the other materials were used as oxygen sensitive materials. The temperature modulation frequency for the different materials was 0.312 Hz. The response time of the sensors was determined by the signal Semiconducting direct thermoelectric gas sensors 381 analyses of thermovoltage and temperature difference. It amounted to 6.4 s. All the sensors showed a reproducible and rapid behavior. For the DTEGs, it was shown that it is possible to mill out a significant part of the gas sensitive layer without affecting the measured thermopower significantly. With an adapted design, DTEGs may be a good alternative to resistive gas sensors. Besides the encouraging results, DTEGs still have great potential for further improvements. The temperature modulation frequency of 0.312 Hz is not sufficient for all applications. For fast responding devices, the temperature modulation has to be in the range of 100 Hz. Manufacturing technology needs to be adjusted to achieve this. Micro- machined ceramics or silicon hot plate gas sensors may be preferred because such a high-temperature modulation frequency needs quite small structures that may not be feasible with conventional ceramic thick-film technology. Until now, only a few materials have been studied for DTEG application. The research focus for gas sensitive materials is usually on resistive materials. The possibility of intrinsic materials with an enhanced sensitivity (when applied in the DTEG mode) is an important property that needs to be addressed in the future. Intrinsic (low conducting) behavior can be improved, when materials with a high mobility of charge carriers are used. Then, films with a low internal resistance and a high sensitivity can be obtained. Furthermore, the aspect of utilizing ion conducting materials should be more emphasized because a high selectivity can be expected because of the distinct ion conduction. References 1. Riegel J, Neumann N, Wiedenmann HM. Exhaust gas sensors for automotive emission control. Solid State Ionics 2002;152:783e800. 2. Moos R. A brief overview on automotive exhaust gas sensors based on electroceramics. Int J Appl Ceram Technol 2005;2:401e13. 3. Denk I, Ingrisch K, Weigold T, Baumann K, Weiblen K, Bauer M, Zeppenfeld A, Schuhmann B. New CO/NOx-sensor system for automotive climate control in a small size housing with high mounting flexibility. Sensor 99 Proc 1999:339e44. 4. Williams DE. Semiconducting oxides as gas-sensitive resistors. Sensor Actuator B Chem 1999;57:1e16. 5. Yamazoe N. Toward innovations of gas sensor technology. Sensor Actuator B Chem 2005;108:2e14. 6. Taguchi N. Gas detecting element and making of it. 1970. US patent specification, US3644795. 7. Logothetis EM, Kaiser WJ. TiO2 film oxygen sensors made by chemical vapour deposition from organometallics. Sensor Actuator 1983;4:333e40. 8. Takami A. Development of titania heated exhaust-gas oxygen sensor. Am Ceram Soc Bull 1988;67(12):1956e60. 382 F. Rettig and R. Moos

9. Gerblinger J, Meixner H. Fast oxygen sensors based on sputtered strontium-titanate. Sensor Actuator B Chem 1991;4(1e2):99e102. 10. Schonauer€ U. Response-times of resistive thick-film oxygen sensors. Sensor Actuator B Chem 1991;4(3e4):431e6. 11. Moos R, Menesklou W, Schreiner HJ, H€ardtl KH. Materials for temperature independent resistive oxygen sensors for combustion exhaust gas control. Sensor Actuator B Chem 2000;67(1e2):178e83. 12. Cantalini C, Pelino M, Sun HT, Faccio M, Cantucci S, Lozzi L, Passacantando M. Cross sensitivity and stability of NO2 sensors from WO3 thin film. Sensor Actuator B Chem 1996;35(1e3):112e8. 13. Lampe U, Fleischer M, Meixner H. Lambda measurement with Ga2O3. Sensor Actuator B Chem 1994;17(3):187e96. 14. Jayaraman V, Gnanasekar KI, Prabhu E, Gnanasekaran T, Periaswami G. Preparation and characterisation of Cr2exTixO3þd and its sensor properties. Sensor Actuator B 1999;55(2e3):175e9. 15. Niemeyer D, Williams DE, Smith P, Pratt KFE, Slater B, Catlow CRA, Stoneham AM. Experimental and computational study of the gas-sensor behaviour and surface chemistry of the solid-solution Cr2exTixO3 (x < ¼ 0.5). J Mater Chem 2002;12(3): 667e75. 16. Ruiz AM, Sakai G, Cornet A, Shimanoe K, Morante JR, Yamazoe N. Cr-doped TiO2 gas sensor for exhaust NO2 monitoring. Sensor Actuator B Chem 2003;93(1e3):509e18. 17. Chou SM, Teoh LG, Lai WH, Su YH, Hon MH. ZnO/Al thin film gas sensor for detection of ethanol vapor. Sensors 2006;6:1420e7. 18. Moos R, Rettig F, Hurland€ A, Plog C. Temperature-independent resistive oxygen exhaust gas sensor for lean-burn engines in thick-film technology. Sensor Actuator B Chem 2003;93(1e3):43e50. 19. Rettig F, Moos R, Plog C. Poisoning of temperature independent resistive oxygen sensors by sulfur dioxide. J Electroceram 2004;13(1e3):733e8. 20. Gerblinger J, Lampe U, Meixner H. German patent specification. 1993. DE4339737C1. 21. Rettig F, Moos R, Plog C. Sulfur adsorber for thick-film exhaust gas sensors. Sensor Actuator B Chem 2003;93(1e3):36e42. 22. Ivers-Tiffee E, H€ardtl KH, Menesklou W, Riegel J. Principles of solid state oxygen sensors for lean combustion gas control. Electrochim Acta 2001;47(5):807e14. 23. Somov SI, Guth U. A parallel analysis of oxygen and combustibles in solid electrolyte amperometric cells. Sensor Actuator B Chem 1998;47:131e8. 24. Baunach T, Sch€anzlin K, Diehl L. Sauberes Abgas durch Keramiksensoren. Phys J 2006; 5(5):33e8. 25. Heikes RR, Ure RW. Thermoelectricity. Interscience Publishers; 1961. 26. Choi GM, Tuller HL, Goldschmidt D. Electronic-transport behavior in single- crystalline Ba0.03Sr0.97TiO3. Phys Rev B 1986;34(10):6972e9. 27. Rettig F. Direkte thermoelektische Gassensoren (in German). PhD thesis, University of Bayreuth, Shaker-Verlag; 2008. 28. Nagy PB, Nayfeh AH. On the thermoelectric magnetic field of spherical and cylindrical inclusions. J Appl Phys 2000;87:7481e90. 29. Shin W, Matsumiya M, Izu N, Murayama N. Hydrogen-selective thermoelectric gas sensor. Sensor Actuator B Chem 2003;93(1e3):304e8. 30. Willet M. Recent developments in catalytic gas sensors. In: 1st international workshop an smart gas sensors technology an application. Germany: Freiburg im Breisgau; 2005. 31. Balducci A, D’Amico A, Di Natale C, Marinelli M, Milani E, Morgana ME, Pucella G, Rodriguez G, Tucciarone A, Verona-Rinati G. High performance CVD-diamond- based thermocouple for gas sensing. Sensor Actuator B Chem 2005;111:102e5. Semiconducting direct thermoelectric gas sensors 383

32. Pisarkiewicz T, Stapinski T. Influence of gas atmosphere on thermopower measurements in tin oxide thin-films. Thin Solid Films 1989;174:277e83. 33. Siroky K. Use of the Seebeck effect for sensing flammable-gas and vapors. Sensor Actuator B Chem 1993;17(1):13e7. 34. Mizsei J. H2-induced surface and interface potentials on pd-activated SnO2 sensor films. Sensor Actuator B Chem 1995;28(2):129e33. 35. Moos R. Method and apparatus for detecting the oxygen content of a gas. US patent specification; 1998US6368868. 36. Ionescu R. Combined Seebeck and resistive SnO2 gas sensors, a new selective device. Sensor Actuator B Chem 1998;48(1e3):392e4. 37. Liess M, Steffes H. The modulation of thermoelectric power by chemisorption: a new detection principle for microchip chemical sensors. J Electrochem Soc 2000;147(8): 3151e3. 38. Smulko JM, Ederth J, Li YF, Kish LB, Kennedy MK, Kruis FE. Gas sensing by thermoelectric voltage fluctuations in SnO2 nanoparticle films. Sensor Actuator B Chem 2005; 106(2):708e12. 39. Keem JE, Honig JM. Seebeck measurements and their interpretation in high-resistivity materials case of semiconducting V2O3. Phys Status Solidi A-Appl Res 1975;28(1): 335e43. 40. Keithley. Low level measurements handbook. Keithley Instruments Inc; 2004. 41. Timm H, Janek J. On the Soret effect in binary nonstoichiometric oxide-skinetic demixing of cuprite in a temperature gradient. Solid State Ionics 2005;176:1131e43. 42. Rettig F, Moos R. Temperature-modulated direct thermoelectric gas sensors: thermal modeling and results for fast hydrocarbon sensors. Meas Sci Technol 2009; 20(6). 065205 9. 43. Rettig F, Moos R. Direct thermoelectric gas sensors: design aspects and first gas sensors. Sensor Actuator B Chem 2007;123(1):413e9. 44. Simon T, Barsan N, Bauer M, Weimar U. Micromachined metal oxide gas sensors: opportunities to improve sensor performance. Sensor Actuator B Chem 2001;73(1):1e26. 45. Kittel C, Kromer€ H. Thermal physics. W. H. Freeman; 1980. 46. Jamnik J, Kamp B, Merkle R, Maier J. Space charge influenced oxygen incorporation in oxides: in how far does it contribute to the drift of Taguchi sensors? Solid State Ionics 2002;150(1e2):157e66. 47. Rettig F, Moos R. Morphology dependence of thermopower and conductance in semiconducting oxides with space charge regions. Solid State Ionics 2008b;179(39): 2299e307. 48. Maier J. Physical chemistry for ionic material: ions and electrons in solids. John Wiley & Sons Ltd; 2004. 49. Tschope€ A. Interface defect chemistry and effective conductivity in polycrystalline cerium oxide. J Electroceram 2005;14:5e23. 50. Henisch HK. Semiconductor contacts. Oxford University Press; 1984. 51. Barsan N. Conduction models in gas-sensing SnO2 layers: grain-size effects and ambient atmosphere influence. Sensor Actuator B Chem 1994;17(3):241e6. 52. Wolkenstein T. Electronic processes on semiconductor surfaces during chemisorption. New York: Consultants Bureau; 1991. 53. Yoo HI, Song CR. Thermoelectricity of BaTiO3þd. J Electroceram 2001;6:61e74. 54. Rothschild A, Menesklou W, Tuller HL, Ivers-Tiffee E. Electronic structure, defect chemistry, and transport properties of SrTi1exFexO3ey solid solutions. Chem Mater 2006;18:3651e9. 55. Moos R, Izu N, Rettig F, Reiß S, Shin W, Matsubara I. Resistive oxygen gas sensors for harsh environments. Sensors 2011;11(4):3439e65. 384 F. Rettig and R. Moos

56. Rettig F, Sahner K, Moos R. Thermopower of LaFe1exCuxO3ed. Conf Proc Solid State Ionics 2005;15:569. Baden-Baden. 57. Rettig F, Moos R. Direct thermoelectric hydrocarbon gas sensors based on SnO2. IEEE Sens J 2007b;7:1490e6. 58. Jonker GH. Application of combined conductivity and Seebeck-effect plots for analysis of semiconductor properties. Philips Res Rep 1968;23(2):131e8. 59. Moos R, H€ardtl KH. Defect chemistry of donor doped and undoped strontium titanate ceramics between 1000 C and 1400 C. J Am Ceram Soc 1997;80:2549. 60. Roder-Roith€ U, Rettig R, Roder€ T, Janek J, Moos R, Sahner K. Thick-film solid electrolyte oxygen sensors using the direct ionic thermoelectric effect. Sensor Actuator B Chem 2009;136(2):530e5. 61. Gurlo A, Barsan N, Oprea A, Sahm M, Sahm T, Weimar U. An n-top-type conductivity transition induced by oxygen adsorption on Fe2O3. Appl Phys Lett 2004a;85(12):2280e2. 62. Gurlo A, Sahm M, Oprea A, Barsan N, Weimar U. A p-ton-transition on Fe2O3-based thick film sensors studied by conductance and work function change measurements. Sensor Actuator B Chem 2004b;102(2):291e8. 63. Rettig F, Moos R. a-iron oxide: an intrinsically semiconducting oxide material for direct thermoelectric oxygen sensors. Sensor Actuator B Chem 2010;145(2):685e90. 64. Ahlgren EO, Poulsen FW. Thermoelectric power of stabilized zirconia. Solid State Ionics 1995;82:193e201. 65. Roder-Roith€ U, Rettig F, Sahner K, Roder€ T, Janek J, Moos R. Perovskite-type proton conductor for novel direct ionic thermoelectric hydrogen sensor. Solid State Ionics 2011;192(1):101e4. 66. Williams D, Tofield B, McGeehin P. Oxygen sensors. European patent specification; 1985. EP00062994. CHAPTER TWELVE

Dynamic operation of semiconductor sensors

Andreas Schutze,€ Tilman Sauerwald Lab for Measurement Technology, Department Systems Engineering, Saarland University, Saarbrucken,€ Germany

Contents

12.1 Introduction 385 12.2 Dynamic operation of metal oxide semiconductor gas sensors 388 12.2.1 Temperature-cycled operation 390 12.2.2 Field effect 396 12.2.3 Optical excitation 398 12.3 Dynamic operation of gas-sensitive ﬁeld-effect transistors 398 12.3.1 Temperature-cycled operation for SiC-FET sensors 399 12.3.2 Gate biasecycled operation 401 12.3.3 Current compensation mode 403 12.3.4 Combined methods 403 12.4 Conclusion and outlook 404 References 408

12.1 Introduction Semiconductor gas sensors offer a range of advantages. Especially their high sensitivity and robust long-term performance combined with low cost make them attractive for various applications. On the other hand, they also pose considerable challenges due to their typically low selectivity and poor stability, i.e., baseline drift and changes in sensitivity. Note that high robustness and poor stability are not always a contradiction. For instance, metal oxide semiconductor (MOS) gas sensors have a proven lifetime of several decades for detection of explosive gas leaks, i.e., high concentrations exceeding typical ambient variations, and for monitoring air quality or more correctly sudden changes in air quality in cars just to mention two main applications. In both cases, the application does not require a stable baseline or constant sensitivity for determination of a gas concentration.

Semiconductor Gas Sensors, Second Edition ISBN: 978-0-08-102559-8 © 2020 Elsevier Ltd. https://doi.org/10.1016/B978-0-08-102559-8.00012-4 All rights reserved. 385 j 386 Andreas Schutze€ and Tilman Sauerwald

Stability problems can be overcome with suitable measurement setups. In physical sensors, differential measurements are often used to suppress temperature cross sensitivity (e.g., force and pressure sensors using Wheat- stone bridges) or determine the sensitivity (e.g., magnetic sensors with built-in excitation coils). In chemical sensors, pellistors and mass-sensitive devices often use the same approach with a differential setup combining one gas-sensitive and one inert transducer. Selectivity, on the other hand, is typically not a problem for physical sensors simply due to the much smaller number of relevant factors influencing the sensor signal (or, more abstract, due to the lower dimensionality of the input space). In chemical sensors, the number of relevant factors (or the dimensionality of the input space) is huge: each molecule is basically an independent factor and even seemingly simple environments like indoor air or breath often contain hundreds of gas components with relevant target gases covering a wide concentrations range from ppb level, e.g., benzene, up to several percent, e.g., humidity. To increase the selectivity of chemical sensor systems, biomimetic approaches are often used to emulate the (mammalian) nose.1 The most important approach is the use of multisensor arrays combining various more or less specific sensors and using pattern recognition methods to interpret the resulting signal pattern.2,3 Note, however, that both the nose and multisensor arrays still have severe limitations: the response is strongly nonlinear, different gases or gas mixtures can lead to identical results and the odor response changes due to accommo- dation, saturation, or poisoning effects. One should also point out that the response spectrum of noses has been adapted by evolution: it tends to blank out molecules such as H2, CO, CH4,orH2O, not because these cannot be detecteddour nose detects other small molecules such as NH3,CH2O (formaldehyde), and especially H2S quite sensitivelydbut probably due to their low specificity or information content. Note that multisensor arrays can come in different forms from actual n physical sensors to a single sensor element with multiple electrodes. Furthermore, this can include electrical multiparameter readout (EMR) methods, i.e., current or voltage sweeps or measurement at different frequencies up to impedance spectroscopy: in these cases, additional information about the nonohmic behavior of the sensor is obtained which can reflect the interaction of the gas atmosphere with the sensor surface and thus provide additional information for gas identification. Many methods are used to interpret the response patterns of multisensor e arrays.4 6 Especially artificial neural networks have been widely used to Dynamic operation of semiconductor sensors 387 emulate the data processing in our brain when recording odors, but in principle all methods used in pattern analysis, e.g., for image interpretation or speech recognition, can be also applied to multisensor arrays. Systematic data evaluation is based on four typical steps: (i) data preprocessing, e.g., to reduce noise or eliminate offset; (ii) feature extraction to extract information from the raw data; (iii) feature selection or more generally dimensionality reduction to limit the data to relevant information; and (iv) classification and/or quantification to interpret the data. In addition, as pattern recognition is based on learning from examples rather than model-based interpretation, validation is required to ensure that patterns are not over- interpreted. While achieving impressive selectivity for a wide range of applications, multisensor arrays often exacerbate the limited stability of gas sensor systems: if a single sensor in an array has changed due to drift or poisoning or is otherwise not in its calibrated state, the pattern interpretation will lead to different, often completely false results. Another very powerful approach to achieve differential measurements is based on dynamic excitation of the sensor, i.e., the sensor is changing from one state to another due to external variations. For chemical sensor, the simplest and already quite powerful approach is to measure the sensor signal with and without the influence of the target gas or gas mixture or at least using a controlled variation.7,8 The time response will provide the relative change of the sensor signal, i.e., the conductance for MOX sensors. In addition, the time constant or slope can be observed which reflects the rate of interaction between sensor and gas and components of gas mixtures can be identified by their different time constants. Furthermore, the initial slope of the sensor response is often linear over a much larger concentration range than the steady-state response as nonlinearity is caused by the limited number of interaction sites on the sensor surface limiting the signal at higher coverage. While this method is quite useful for lab evaluation of sensors, i.e., to elucidate the interaction mechanism, its use for practical application is limited due to the requirement of a reference or zero air leading to more complex and costly systems. Note that this approach is also inspired by nature to emulate the sniffing of dogs where the dynamic also provides additional information. Although this method, also referred to as transient response or breathing mode,9,10 has gained increasing interest, this chapter will focus on dynamic sensor operating modes that can be directly controlled electronically to allow full control over the operating mode and make use of the information gain in signal evaluation. 388 Andreas Schutze€ and Tilman Sauerwald

The first, but still most relevant and widely studied, approach for e dynamic operation is temperature modulation,11 13 also referred to as temperature-cycled operation (TCO) due to its inherent repetitive nature. Temperature as the most important physical parameter influencing chemical interaction is a natural candidate for dynamic operation of chemical sensors. While it can in principle be applied to any chemical sensor principle, MOX sensors are prime candidates due to their operation at elevated temperature using integrated heaters. It has also been shown for pellistor-type sensors,14,15 and gas-sensitive field-effect transistors (GasFETs),16 especially those based on SiC due to their operation at elevated temperatures.17 Other e dynamic operating modes are based on the field effect/polarization18 20 and on optical excitation.21,22 Note that, as for the temperature, these parameters are most often used to simply determine the optimal operating point for static operation, but all actually change the equilibrium on the sensor surface and thus provide additional information especially during nonequilibrium states. The general approach is discussed using various terms, i.e., modulation,12,13 transient analysis,23,24 dynamic response,25,26 programming,27 or pulsed operation.28,29 We prefer the term dynamic operation as this clearly implies the active variation of a control parameter (temperature, bias voltage, illumination, etc.) by the sensor electronics allowing application-specific optimization of the sensor system performance. The following sections will discuss dynamic operating modes for MOX and GasFET sensors to show that not only selectivity can be improved but that stability and even sensitivity benefit from this approach.

12.2 Dynamic operation of metal oxide semiconductor gas sensors MOS gas sensors detect redox reactions of gases on the semiconductor surface. These can be direct reactions of the gas with surface states (EleyeRideal mechanism) or indirect reactions requiring the adsorption of the gas followed by a subsequent reaction (LangmuireHinshelwood mechanism). Any redox reaction causes a change in surface charge and therefore in the conductance of the sensor. To understand the impact of surface charge on the conduction, the conduction mechanism needs to be discussed. In general, a surface charge has to be compensated by an opposite charge in the semiconductor. In the case of n-type semiconductors, e.g., tin dioxide, SnO2, the most widely used material for MOX sensors, negative surface charges will be compensated by a positive space charge layer. Dynamic operation of semiconductor sensors 389

The space charge is accompanied by an electrostatic bending of all electronic bands and especially a bending of the conduction band. The height of the band bending Vs and the corresponding energy Eb ¼ qVs can be calculated by Poisson’s equation as a function of the density of negative surface charge Ns, the density of positive charges in the semiconductor given by the number of (ionized) donors Nd, the permittivity of the semiconductor εr; the dielectric constant ε0, and the elementary charge q (Eq. 12.1). 2 2 q Ns Eb ¼ (12.1) 2εrε0Nd Please note that Eq. (12.1) was derived as 1D solution (plane surface) using the Schottky approximation. Because of the band bending, the number of electrons at the surface ns is significantly reduced; it can be estimated by a Boltzmann equation based on the implicit assumption that all donors are ionized (Eq. 12.2). Eb ns ¼ Nd exp (12.2) kbT The conductance model for these surface effects obviously depends on the morphology of the sensor film. For common sensors with a granular thick film, it is often assumed that the conductance is completely dominated by the surface charge and the resulting energy barrier. Therefore, the conductance (Eq. 12.3) can be modeled by a single grain to grain e contact.30 32

Eb G ¼ G0$e kbT (12.3) here G0 denotes the conductance of the granular ﬁlm for the (hypothetical) ﬂat band case. Despite the large variety of possible reaction processes on the surface, the surface coverage in an equilibrium can typically be calculated by a mass action law. An important case, the reaction of reducing gases on an e MOX sensor surface, has been studied intensively.32 34 In this case, a reducing gas R reacts (Eq. 12.4) with ionosorbed oxygen, which is itself adsorbed at a surface site s (Eq. 12.5), in a direct reaction. þ # þ Ox R ROx e (12.4)

1 s þ O þ e #O (12.5) 2 2 ads 390 Andreas Schutze€ and Tilman Sauerwald

Of course, Eqs. (12.4) and (12.5) only show a simpliﬁed reaction scheme for a direct reaction of the reducing gas with a single type of ionosorbed oxygen. A more general discussion of reaction schemes can be found in the studies by Ref. 33 and Ref. 34. Moreover, even an additional electron transfer to the semiconductor yielding an accumulation layer has been reported at high excess of reducing gas at the surface.35,36 However, in most cases, i.e., in air with only a small concentration of reducing gas, the number of free electrons at the surface is limiting the adsorption of the oxygen, which is itself dependent on the band bending caused by oxygen adsorption. This limitation, also known as Fermi level pinning, is the reason that the actual density of adsorbed oxygen is only changing very little at the equilibrium of reactions (12.4) and (12.5).On the other hand, even small changes in surface charge cause quite large changes in the MOS sensor conductance (cf. Eq. 12.3). The sensor response Seq (the subscript eq is indicating the equilibrium condition) deﬁned as the change of conductance given by the conductance under reducing gas Ggas divided by the conductance in air Gair can be estimated in equilibrium condition by a power law.32,33

¼ Ggas ¼ n Seq 1 kcgas (12.6) Gair The exponent n is equal to 0.5 for the basic reaction scheme shown in Eqs. 12.4 and 12.532 and can have other rational values for other, more complex reaction schemes.33,34 12.2.1 Temperature-cycled operation It is obvious that temperature has a strong impact on numerous effects in MOX sensors, among others on the rate constants of the redox reactions. For most redox processes, an activation energy needs to be overcome before the reaction can take place. Gases can therefore be discriminated with respect to their activation energy. In 1974, the principle was utilized for the first time in a procedure to selectively detect carbon monoxide and hydrocarbons, mainly methane, in firedamp.11 In the following years, e many studies using TCO were published12,25,37 41 reporting the use in various application fields such as fire detection, air quality sensing, and in the detection of emissions. In an earlier work, we have demonstrated the selectivity achieved with TCO41 for detection of three hazardous VOCs (volatile organic compounds), benzene, formaldehyde, and naphthalene, at very low concentrations. The compounds were tested at two Dynamic operation of semiconductor sensors 391 concentrations, one below and one above the recommended threshold limit values in air: the tested concentrations were 0.5 and 5 ppb for benzene, 10 and 100 ppb for formaldehyde, and 2 and 20 ppb for naphthalene. All concentrations were tested at various ambient humidity levels and in the presence of 0, 500, or 2000 ppb ethanol as a typical, nonhazardous VOC. Three different MOS sensors (GGS 1330, GGS 2330, and GGS 5330, all from UST, Umweltsensortechnik, Geschwenda, Germany) have been used each with a specific temperature cycle. For each sensor, a set of 40 features were calculated from the sensor response during the temperature cycle, describing the shape of sensor conductance. Using these shape- describing features, a linear discriminant analysis (LDA) was performed. LDA is a supervised training method that maximizes the distance between different groups with respect to the scattering within the groups.4 The resulting LDA projections for the each of the three sensors and for the data fusion of all sensors are shown in Fig. 12.1. The discrimination of the three toxic gases from each other and from air is quite successful for all three sensors, but there is still some overlap. For the validation of the results, leave-one-out cross validation (LOOCV) is performed4 indicating in all cases a correct classification of over 95% of the measurements. Data fusion of the three sensors improves the discrimination further (Fig. 12.1, lower right). No overlap is observed even in the 2D-scatterplot and, consequently, LOOCV results in 100% correct classification. Many investigations use a somewhat haphazard way of defining the temperature profile of TCO rather than a consistent optimization process, due to the numerous effects that make the derivation of a universal model very challenging. However, in the last few years, the advantages of a model-based optimization became obvious. To allow an objective comparison of various temperature profiles, the definition of the sensor response is extended to the concept of a quasi-static sensor response SqsðtÞ at a well-defined time t within each temperature cycle (Eq. 12.7).

GgasðtÞ SqsðtÞ¼ 1 (12.7) GairðtÞ It was reported that the quasi-static sensor response can in some cases be estimated by a power law similar to the equilibrium condition,42,43 in some cases showing a large improvement of the sensor response compared with operation at constant temperature. In other cases, the quasi-static sensor response has a completely different characteristic with the gas concentra- tion44,45 due to the fact that the surface coverage in the TCO can be far 392 Andreas Schutze€ and Tilman Sauerwald

Figure 12.1 LDA projection for the discrimination of hazardous VOCs in air (at various humidities and against changing ethanol background up to 2 ppm). (a) data of GGS 1330 sensor; (b) data of GGS 2330 sensor; (c) data of GGS 5330 sensor; (d) fusion of all sensor data (modified after [41]). from equilibrium. Looking at the numerous effects of temperature on the sensor conductance, we start with the fact that the conductance is thermally activated (Eq. 12.3). As this is due to the statistic distribution of the conduction band electrons, this thermal activation is following any temperature variation almost immediately. More complicated are the redox processes on the surface, which need significant time to reach an equilibrium due to their activation energy. Nakata et al. have proposed that the reactions should then be expressed by several temperature-dependent rate constants26,46,47 which should be considered in the evaluation of gas sensor. To this end, they presented an FFT (fast Fourier transformation)-based method for feature extraction, which should implicitly reflect the different rate constants. An explicit determination of the reaction rates was enabled with the introduction of micromachined membrane sensors, which allow heating Dynamic operation of semiconductor sensors 393 and especially cool down of the sensing layer on a very short time scale of a few milliseconds. Using these devices, Ding et al.48 proposed and tested a set of rate equations for the modeling of the surface charge, which in principle allow the modeling of the temperature cycle. Following this approach, Baur et al. developed a method to optimize the sensor response within TCO using a rate equation model.44,49 In this model, only a single negatively charged surface state is assumed to simplify the rate equations to one adsorption and one desorption term (Eq. 12.8). dNs xDH ¼ kans exp kdNs (12.8) dt 2kbT The adsorption is determined by the concentration of electrons and the enthalpy of adsorption DH for the negative charge (ionosorbed oxygen OÞ and by a rate constant ka for adsorption. The desorption is given by the density of surface charge Ns and a rate constant kd. The increase of sensor response is because the activation energy EbNðTÞ caused by the band bending (Eq. 12.1) in equilibrium is strongly temperature dependent. A strong increase of EbNðTÞ and therefore of NsNðTÞ was observed with temperature (Fig. 12.2). A high temperature period in the temperature cycle can therefore provide a surface with a large excess of surface oxygen, which will cause a predominant reduction of the sensor accompanied by strong sensor response during any following low temperature period. The principle of this mode of

0.9

0.8

0.7

0.6

0.5

Activation energy (eV) 0.4

0.3

0.2 100 150 200 250 300 350 400 450 Temperature (°C) Figure 12.2 Activation energy in equilibrium over temperature (modiﬁed after [49]). 394 Andreas Schutze€ and Tilman Sauerwald

– 0x

dlnG Eb,Tlow, ∞ 1 = – d( T) kB dEB Δt ∞ 0 – Δ x dt – t 0 0x

Δt 0 ∞ Δt dEB E ∞

Logarithm of conductance dlnG b,Thigh, dt 1 = – d( T) kB – 0x

Inverse temperature Figure 12.3 Principle of a TCO for the optimization of the sensor response (modiﬁed after [44]). The light grey area represents the electron-depleted region at the grain surface. The chemisorbed oxygen is represented by dashes. operation is shown in Fig. 12.3. The operating mode is divided into four sections (ieiv): (i) Equilibration during high temperature period to achieve a high negative surface charge (oxidation). The energy barrier increases with the surface coverage according to Eq. (12.1). The equilibrium

level is then denoted as Eb;Thigh;N. (ii) Rapid decrease of temperature to preserve excess surface charge. The logarithmic conductance decreases with a constant slope due to its thermal activation (Arrhenius’ law). (iii) Detection period at low temperature with predominant surface reduction. The change of the energy barrier is detected with respect

to the gas concentration (the new equilibrium is denoted as Eb;Tlow;N, but this equilibrium does not necessarily have to be reached within the temperature cycle). (iv) (Rapid) increase of temperature to restart oxidation period. Because of the high excess of ionosorbed oxygen, the rate equation at the beginning of the low temperature period (iii) can be simpliﬁed to Eq. 12.9. X dNs ¼ ¼ þ j kdNs with kd kdes kred (12.9) dt j Dynamic operation of semiconductor sensors 395

The total rate constant kd contains all mechanisms for surface reduction, i.e., the desorption of ionosorbed oxygen kdes and the redox reactions of the j ionosorbed oxygen kred with different reducing gases j. In this mode, the sensor response can be very high. In fact, for 1 ppm ethanol in air, we observed an increase of the response compared with isothermal operation by a factor of 1000.49 Note that this boost in sensitivity is strongly depending on the gas concentration as can be observed by comparing the pulse peak height and the near steady-state response at the end of the plateaus in Fig. 12.4. For selective detection of gases, i.e., the original use of TCO, the temperature cycle has to combine various “low” temperature periods j at different temperatures to make use of the change of kred with temperature for gas identiﬁcation (Fig. 12.4). The sensor response shows a distinct maximum on each plateau. For ethanol, which is easily oxidized, the response is highest at the lowest operating temperature of 130 C, whereas for benzene, a relatively stable molecule, the highest response is observed between 220 and 270 C. At the beginning of the low temperature period, the sensor response is obviously no longer following a power law, as the surface coverage decreases linearly with the applied gas concentration (Eq. 12.9).50 Together with Eqs. (12.1) and (12.3), this leads to Eq. (12.10) showing the exponential dependency of the sensor response with respect to k j : 0 1red ð Þ X ð Þ¼ exp 2 kd ti ¼ @ j A Sqs ti ð Þ exp kred ti (12.10) exp 2 kdes ti j

Figure 12.4 Sensor response to ethanol (solid lines) and benzene (dashed) at various concentrations during the temperature cycle shown at the bottom (modiﬁed after [49]). 396 Andreas Schutze€ and Tilman Sauerwald

Please note that Eq. (12.10) uses the approximation of negligible changes of Ns (cf. Eq. 12.9) and is therefore a valid approximation only for a short period after the temperature change. A more generalized method for the determination of the rate constants and their use as sensor signal can be found in Ref. 50. For a direct reaction of the reducing gas (EleyeRideal j j j mechanism), kred is proportional to the gas concentration cgas (a being a proportionality constant) and therefore the sensor response is simply given by 0 1 X ð Þ ¼ @ j j A Sqs ti exp a cgas ti (12.11) j For a LangmuireHinshelwood reaction mechanism, this dependency might be more complex depending on the adsorption isotherm of the reducing compound. For low concentrations, however, a linear isotherm is often appropriate (Henry isotherm). In this case, Eq. (12.11) is also applicable for this type of reaction. Obviously, this approach to TCO optimization is especially suitable for selectively measuring small concentrations of reducing gases, which is, e.g., required for determination of air quality. For the detection of benzene in pure air, this method has been successfully tested yielding very good results. In this experiment, three low temperature plateaus were used and on each the rate constant kdðTÞ was calculated as model-based feature, which were then used as (nearly linear) virtual sensors in a multilinear quantiﬁcation algorithm (partial least square regression) as shown in Fig. 12.5. With small variations, Eq. (12.11) can also be used for the description of changing gas concentrations, e.g., for gas pulses.52 (Please note that for clarity only a single gas component is represented in Eq. 12.12.) 0 1 Zt @ 0A SqsðtÞ ¼ exp a cgas dt (12.12) 0 In this sense, the principle of temperature modulation can also be used for systems with discontinuous gas application like sensor preconcentrator or gas chromatographic systems.52,53 12.2.2 Field effect A dynamic change of surface charge and surface reactions can also be obtained through the variation of electrical ﬁelds. An external electrical Dynamic operation of semiconductor sensors 397

Training date Training date 10 Regression 10 Test date Regression

8 8

6 6

4 4

2 2 Sensor system readout / ppb Sensor system readout / ppb

0 0

0 2 4 6 8 10 0 2 4 6 8 10 Concentration set point /ppb Concentration set point /ppb Figure 12.5 Quantiﬁcation of benzene using a three-sensor array with response optimized temperature cycles for the detection of benzene at (sub-)ppb level. The quantiﬁcation was performed using the pre-processed sensor data (rate constants) and PLSR. On the left: training data with 10% and 40% RH; on the right: training and test data including various untrained benzene concentrations and all benzene concentrations at 25 % RH.51

field, e.g., caused by a suspended electrode, will cause compensating charges on the sensor54 which can favor the adsorption or desorption of ionosorbed species such as NO2 depending on the polarity of the electrode. Dynamic variation of perpendicular electrical fields is mostly used in GasFET setups (cf. Section 12.3.2) as the gate electrode provides an easy access to this parameter. However, in some types of MOS sensors, a simple modulation of the electrical field for the sensor readout can be used to obtain several e virtual sensors.55 57 Tungsten oxide can often contain mobile donors, which can be polarized even with small electrical fields. The concentration of the donors is directly linked to the band bending (at constant surface charge) according to Eq. (12.1). Donor-accumulated regions therefore initially show a smaller band bending and a lower resistance. After the donor accumulation and the change in band bending, a change in oxygen surface coverage according to Eq. (12.2) would be expected with a certain increase of the band bending. Using a four electrode tungsten oxide sensor, this accumulation and surface relaxation has been observed by a decrease and subsequent increase of the resistance (as well as the expected inverse behavior at the donor depletion region).56 It was demonstrated that a cyclic variation of readout voltages can even be used to generate gas-specific signals, i.e., to boost selectivity, as shown in Fig. 12.6.57 398 Andreas Schutze€ and Tilman Sauerwald

Figure 12.6 Two virtual sensors obtained by feature extraction of resistance change during a voltage pulse. The total duration of the bias cycle is 300 s with two opposite voltage pulses of 600 mV for 30 s each followed by 120 s relaxation periods [58].

12.2.3 Optical excitation Optical excitation induces a broad variety of reactions that can be used for dynamic operation. Irradiation with UV light typically causes the generation of electronehole pairs that can act as counterparts for a redox reaction at the surface. Especially holes, which are typically not abundant in n-type semiconductors, play an important role in the recombination with ionosorbed, oxidized species such as NO2 or O3. Thus, optical excitation is often used to activate the desorption process for improving the sensor kinetics at low temperatures and even at room temperature for various sensor materials e ranging, e.g., from tin oxide,58 titanium oxide,59 and indium oxide60 63 to organic semiconductors.64 However, only very few authors actually use optical excitation to create dynamic signals.61,62,64 When used in dynamic operation, the optical excitation was shown to have a signiﬁcant potential for selective detection of gases.64

12.3 Dynamic operation of gas-sensitive field-effect transistors Gas-sensitive field-effect devices, especially transistors (GasFETs), offer additional modes of interaction between the semiconductor sensor Dynamic operation of semiconductor sensors 399 and the atmosphere. While the signal in MOS sensors is only resulting from charge transfer between gas and surface (chemisorption) or redox reactions on the surface, i.e., chemical processes, GasFETs can also detect gas due to physisorption due to the field effect of polar molecules. The first GasFETs demonstrated by Lundstrom€ 65 with solid palladium gates were only able to detect hydrogen and hydrogen-containing gases due to ionized hydrogen, i.e., protons, diffusing through the gate. The response and application spectrum were greatly expanded with novel technological approaches such as suspended or perforated gates and by making use of silicon carbide (SiC) as semiconductor instead of Si. SiC allows operation at elevated temperatures due to its much larger bandgap as well as in harsh environments due to its chemical inertness (cf. Chapter 8). However, with these approaches, GasFETs face similar challenges as MOS sensors, i.e., limited selectivity and stability due to the broad response spectrum and drift or poisoning effects caused by slow or irreversible chemical processes on the surface. On the other hand, GasFETs offer additional potential for dynamic operation: not only temperature modulation for SiC-FET sensors similar to MOX sensors but also variation of the gate bias VGS as a direct way to influence the response spectrum and sensitivity of GasFET sensors as shown in Fig. 12.7.16 While dynamic modes of operation have been widely utilized for MOX sensors already for over 40 years,11 first results utilizing temperature and gate bias modulation for GasFETs were published around 200066,67 and 2010,16,20 respectively. Both methods, individually and in combination, were systematically studied by C. Bur.17 12.3.1 Temperature-cycled operation for SiC-FET sensors Similar to MOS sensors, TCO has proven to be a powerful tool for improving the selectivity of SiC-FET sensors.68,69 Fig. 12.8 shows one example for the discrimination and quantification of hazardous VOC (benzene, formaldehyde, and naphthalene), e.g., for indoor air quality control and demand-controlled ventilation.70 Other potential applications include exhaust gas monitoring, i.e., selective quantification of NO and/ 68,69 or NH3 for SCR systems in diesel engines or for SO2 in power plant monitoring and control.71,72 The principal processes on the sensor surface are similar to those occurring on MOS sensors, i.e., temperature-induced changes in ionosorption of oxygen and other redox species as well as adsorption of polar molecules. However, the mechanism cannot be modeled today with a simple approach 400 Andreas Schutze€ and Tilman Sauerwald

(a)

D S n-type active layer p-type buffer layer VDS VGS n-type 4H-SiC substrate

(b) 200 °C Background0 500 NH3 (500 ppm) CO (500 ppm) V =2V 400 GS

300 A) μ (

DS VGS=0V I 200

100

V =–2V 0 GS

20 % O2,0%r.h. 012345

VDS (V) Figure 12.7 (a) Schematic cross-section of a SiC-FET; (b): typical IV-curves of the SiC-FET in pure dry air (black) and with 500 ppm ammonia (NH3, orange, dashed) and carbon monoxide (CO, green, dashed) for different values of the applied gate potential VGS.In static mode, this can be used to maximize the sensor response (VGS ¼ 2 V) or to improve selectivity (here: NH3 vs. CO at VGS ¼ 0 V) (modiﬁed after [16]). as for the MOS sensors, cf. Section 12.2.1. This is due, on the one hand, to the more complex structure of the sensor surface with catalyst clusters, open insulator surface, and the highly relevant three-phase boundaries between catalyst, surface, and atmosphere.73 On the other hand, the much larger thermal time constants of SiC-FET sensors (several seconds compared with approximately 10 ms for microstructured MOX sensors) do not allow experimental differentiation between temperature changes and the induced gas adsorption and redox reactions. Dynamic operation of semiconductor sensors 401

(a) S (c) A Naphthalene S Benzene T

Below threshod D F Formaldehyde N B

(b) (d)

B A C

F T

D N B T

T Figure 12.8 (a) temperature cycle, actual temperature and resulting sensor signal

VDS (constant current mode: IDS ¼ 45 mA) in air for a SiC-FET (cf. Fig. 12.7); (b) normalized difference signal of the sensor response for 100 ppb formaldehyde, 20 ppb naphthalene, and 4.5 ppb benzene at 40 % RH in air; (c) discrimination of benzene, naphthalene, and formaldehyde based on LDA: each group contains three different concentrations above the ventilation threshold of the respective gas; the fourth group contains pure air as well as one concentration of each gas below the threshold. All groups contain data at 20 % and 40 % RH. Using a Mahalanobis distance classifier the achieved leave-one-out cross-validation rate of nearly 90 % indicates the almost perfect discrimination of all gases and background. (d) Quantification of naphthalene based on LDA: the plot shows gas concentration vs. value of first discriminant function with a second order fit for the training data (0, 5 and 40 ppb, solid symbols); evaluation data (2.5, 10 and 20 ppb) are marked by open symbols. The indicated line could be used as a threshold limit value, e.g. for ventilation control (modified after [71]).

12.3.2 Gate biasecycled operation Typical measurements of GasFETs are performed by setting two electrical parameters of the transistor, e.g., the gate bias VGS and the drain-source voltage VDS (cf. Fig. 12.7), at constant values and measuring the resulting 402 Andreas Schutze€ and Tilman Sauerwald

(a) (b)

(c)

Figure 12.9 Hysteresis of DIDS ¼ IDS(gas) - IDS(background) for CO (green, triangles) and NH3 (orange, circles) in pure nitrogen at 50 % RH at 187 C (a) and 265 C (b) and in air at 50 % RH at 265 C (c). The shape of the hysteresis is greatly inﬂuenced by the background gas, the humidity and the sensor temperature; the observed cross-over points indicate multiple processes taking place on and in the sensor (modiﬁed after [16]).

third parameter, in this case the drain-source current IDS. The control parameters are selected to optimize the sensitivity or the selectivity versus relevant interfering gases. Additional information is obtained by dynamically changing one control parameter, e.g., the gate bias, during the measurement to probe different operating points of the transistor, resulting in a measurement signal over the control parameter cycle, i.e., gate biasecycled operation, Fig. 12.9. One seemingly obvious advantage would be the possibility to use faster cycles by employing fast changes of the electrical control parameter. We found, however, that the changes induced on the Dynamic operation of semiconductor sensors 403 sensor surface by variation of the gate bias lead to very slow equilibration with time constants in the order of several 10 minutes16 The resulting complex hysteresis curves, Fig. 12.9, indicate multiple processes competing on and in the sensor, i.e., ionosorption on catalyst and insulator, spill-over from catalyst to insulator andddue to the extremely long time constantsdprobably diffusion of ions into the sensor layers. This latter assumption is also corroborated by the inﬂuence of the insulator material on the sensor performance.74 Gate bias cycling therefore does not only allow a signiﬁcant increase in the selectivity16 but can also be applied for experimental studies to achieve a better understanding of the relevant processes determining the sensor behavior and, thus, for improving the sensor performance further. 12.3.3 Current compensation mode

Another approach for dynamic operation is achieved by keeping both VDS and IDS constant by a closed-loop control of the gate bias VGS. In effect, this means that gas-induced changes on the sensor surface are compensated by a change of the gate bias. This method was recently tested in a preliminary study which found that the signal to noise ratio is only slightly decreased in this operating mode if suitable electronics with sufficiently high resolution are used for measuring the current and setting the gate voltage.75,76 Again, this method can be used to improve our understanding of the processes on the sensor surface as, ideally, the resulting voltage difference DVGS directly reflects the change of the control voltage of the transistor, i.e., the effective charge caused by the additional gas adsorption on the surface allowing direct comparison of different gases, gas mixtures, and concentrations. In addition, the current compensation mode proved to result in a more linear sensor response allowing quantification of ammonia in the range from 0 to 30 ppm with constant uncertainty as shown in Fig. 12.10. 12.3.4 Combined methods Dynamic operating modes can be combined to achieve even better performance for the overall system. The results shown in Fig. 12.10 are actually obtained by combining a simple temperature cycle (linear increase from 180 to 270 C in 25 s followed by a linear temperature decrease back to 180 C in 35 s), thus combining TCO and CCM. For this example, TCO primarily achieves the desired selectivity for discriminating different gases, while CCM improves the quantification due to the improved linearity of the response. 404 Andreas Schutze€ and Tilman Sauerwald

(a) Standard operation 15 6

10 R2 =0,995 log 4

R2 =0,976 5 lin 2 Sensor response / µA rel. sensor response / % 0 0 0102030 Concentration / ppm

(b) Current compensation mode 0.2 8 0.15 6 0.1 4 R2 =0.994 0.05 lin 2 Sensor response / V rel. sensor response / % 0 0 0102030 Concentration / ppm

Figure 12.10 Quantiﬁcation of NH3 with a porous Pt-gate SiC-FET sensor comparing standard operation (a) to current compensation mode (CCM, b); while standard operation results in a logarithmic calibration curve, CCM achieves a linear response and higher relative signals for high concentrations (modiﬁed after [77]).

Similarly, TCO and GBCO can be combined resulting in improved performance as demonstrated for the discrimination of carbon monoxide (CO), nitrogen dioxide (NO2), and ammonia (NH3) independent of concentration and also for quantification of the individual gases.77 Fig. 12.11 demonstrates that a combined cycle achieves better results than TCO or GBCO separately for CO quantification. In the same study, we could also show that, while there is some drift of the sensor, suitable features can be identified allowing stable gas discrimination and quantification by combining calibration data from different states of aging using multivariate statistics.

12.4 Conclusion and outlook Dynamic operation is a powerful and versatile tool for improving the performance of semiconductor gas sensor systems with respect to the three Dynamic operation of semiconductor sensors 405

10 8 Features T-cycle 6 parts A and C 4 2 0 -2 -4 -6 Validation: 99.07 % -8 Wilks' Lambda:0.0157 -10 8 Features 6 GB-cycle @200ºC 4 part D 2 0 -2 -4 -6 Validation: 98.14 % -8 Wilks' Lambda: 0.0082 10 8 Validation: 98.14 % 6 Wilks' Lambda: 0.0023 0 ppm 4 2 200 400 0 -2 600 800 -4 ppm Features -6 T-cycle & GB-cycles -8 parts A to D -10 -15 -10 -5 0 5 10 15 20 25 30 35 40 1st discriminant function Figure 12.11 LDA showing CO quantiﬁcation using features from temperature cycling in the temperature range 200 e 260 C only at VGS ¼ 0 (top), from gate bias variation (-1 to þ2 V) only at 200C (centre) and from the combined cycle (bottom). While the GB-cycle data yield better class separation, T-cycling achieves a higher leave-one-out cross-validation rate. The best results are achieved by combining the features [78].

S, sensitivity, selectivity, and stability. Combined with novel materials based on nanotechnologies and novel substrates/transducers realized with micro- technologies, this approach is one key for addressing new applications with low cost gas sensors, e.g., in environmental monitoring, health, safety, and security. Compared with conventional multisensor arrays, which require 406 Andreas Schutze€ and Tilman Sauerwald frequent recalibration due to their limited stability, the dynamic operation approach provides much better stability due to the inherently differential nature of the measurement comparing different sensor states instead of measuring one specific state only. On the other hand, many similarities are obvious, especially in signal processing: both methods provide multiple raw measurement values which are interpreted with suitable pattern recognition methods. Due to this similarity, the term virtual multisensor was coined to emphasize the fact that a single physical sensor provides the information. Note, however, that this term does not only apply to dynamic sensor operation but also to EMR methods like electrical impedance spectroscopy (EIS). All virtual multisensor methods require advanced electronics to make full use of their potential, both to control the dynamic excitation and to readout the sensor response with the required electrical and temporal resolution. Note that this means that virtual multisensor systems are not necessarily cheaper than multisensor arrays, because the increased cost of the electronics can greatly outweigh the reduced cost of a single versus multiple sensor elements. On the other hand, dynamic operation is more versatile as the operating mode (in addition to the data analysis) can be adapted to different application requirements, in effect shifting from hardware to software solutions. This makes dynamic operation very interesting for novel “generic” application scenarios like gas sensors integrated into smartphones: the target application can be set via software (“there’s an app for that”) allowing different use cases to be addressed with the same hardware. The primary benefit of dynamic operation generally lies in the acquisition of characteristic time constants in addition to the obtained electrical values, which provide further information about the ambient atmosphere. This would seem to suggest that dynamic operation would result in slower response and recovery times for the gas sensor system. However, the opposite is actually the case as could be demonstrated for an application discriminating fuel vapors to prevent false fueling of cars. While the sensor signal of an isothermally operated MOS sensor requires several 10 s to reach steady state in gasoline or diesel vapor atmosphere, the shape of the TCO sensor response could already be evaluated after one temperature cycle with a duration of only 2 s.78 Note that this also allows changing the temperature cycle during the sensor operation, i.e., to improve the classification performance after a first classification step.37 Of course, the achievable measurement rate depends on the time constants to be measured, in this case the interaction between gas molecules and gas-sensitive layer. Dynamic operation of semiconductor sensors 407

This would suggest to use increased temperatures to speed up the chemical processes to achieve a higher measurement rate. However, the temperature also changes the dominating processes on the sensor surface, which limits the potential for tuning the response times with the operating temperature to match the acquisition electronics. Note that EMR methods allow much faster operation (e.g., Fourier-based impedance spectroscopy achieving a complete spectrum over a wide frequency range in only 16 ms79), but obtain only steady-state information and therefore provide less information about the ambient. For instance, interaction of methane (CH4) with a semiconductor surface will be minimal at low temperature due to the high reaction enthalpy and thus identification and quantification will be quite difficult, while for CO the opposite is true. This suggests that combinations of different methods are very attractive to operate the sensor under the best conditions for the required information and/or to extract as much information as possible. Combined operation has been demonstrated for EIS/TCO for MOX sensors79 and for GBCO/ TCO77 as well as CCM/TCO76 for GasFETs. Note that a combination of complementary methods can also allow sensor self-monitoring: by evaluating the sensor response separately with two complementary methods, a sensor that is no longer performing as calibrated can be identified by different predictions resulting from the two methods.6 This can provide a simple and cost-efficient alternative to regular field tests of sensor systems and might be a key for acceptance of semiconductor sensor systems in safety relevant applications like fire or gas leak detection. Finally, novel research results propose to manipulate the gas supply to the sensor with micro preconcentrators (mPC) integrated with gas sensor elements in a microcontainment.53 While gas adsorption/desorption is controlled by the mPC temperature similar to standard sampling and thermal desorption techniques, the gas transport is based on diffusion only thus avoiding mechanical pumps and valves. Because of the compatibility of the mPC and sensor technologies, both being based on microhotplates, this greatly expands the performance spectrum of the integrated gas sensor system. Benefits are the increase of the sensitivity by the increased target gas concentration in the release peak and the improvement of selectivity by suppressing permanent gases, especially H2 and CO, which are not adsorbing on the mPC. Moreover, these systems are also offering improved stability by providing an internal zero air reference: after the release peak, the target gas concentration drops to practically zero in the microsystem as nearly all gas molecules adsorb on the mPC.80 408 Andreas Schutze€ and Tilman Sauerwald

References 1. Schiffman SS, Pearce TC. Introduction to olfaction: perception, anatomy, physiology, and molecular biology. In: Pearce TC, Schiffman SS, Nagle HT, Gardner JW, editors. Handb. Mach. Olfaction e Electron. Nose Technol. Wiley-VCH; 2002. 2. Persaud K, Dodd G. Analysis of discrimination mechanisms in the mammalian olfactory system using a model nose. Nature 1982;299:352e5. https://doi.org/10.1038/ 299352a0. 3. Persaud KC. Electronic gas and odour detectors that mimic chemoreception in animals. Trends Anal Chem 1992;11:61e7. https://doi.org/10.1016/0165-9936(92)80079-L. 4. Gutierrez-Osuna R. Pattern analysis for machine olfaction: a review. IEEE Sens J 2002; 2:189e202. https://doi.org/10.1109/JSEN.2002.800688. 5. Marco S, Gutierrez-Galvez A. Signal and data processing for machine olfaction and chemical sensing: a review. IEEE Sens J 2012;12:3189e214. https://doi.org/ 10.1109/JSEN.2012.2192920. 6. Reimann P, Schutze€ A. Sensor arrays, virtual multisensors, data fusion, and gas sensor data evaluation. In: Wagner T, Kohl CD, editors. Gas Sens. Fundam. Springer Ser. Chem. Sensors biosensors, vol. 15; 2014. 7. Matsunaga N, Sakai G, Shimanoe K, Yamazoe N. Diffusion equation-based study of thin film semiconductor gas sensor-response transient. Sensor Actuator B Chem 2002; 83:216e21. https://doi.org/10.1016/S0925-4005(01)01043-7. 8. Schutze€ A, Pieper N, Zacheja J. Quantitative ozone measurement using a phthalocya- nine thin-film sensor and dynamic signal evaluation. Sensor Actuator B Chem 1995;23: 215e7. https://doi.org/10.1016/0925-4005(94)01281-L. 9. Helwig A, Beer S, Muller€ G. Breathing mode gas detection. Sensor Actuator B Chem 2013;179:131e9. https://doi.org/10.1016/j.snb.2012.07.088. 10. Helwig A, Muller€ G, Sberveglieri G, Faglia G. Gas response times of nano-scale SnO2 gas sensors as determined by the moving gas outlet technique. Sensor Actuator B Chem 2007;126:174e80. https://doi.org/10.1016/j.snb.2006.11.032. 11. Eicker H. Method and apparatus for determining the concentration of one gaseous component in a mixture of gases. 1977. US patent US4012692A, http://www.google.tl/patents/ US4012692. 12. Lee AP, Reedy BJ. Temperature modulation in semiconductor gas sensing. Sensor Actu- ator B Chem 1999;60:35e42. https://doi.org/10.1016/S0925-4005(99)00241-5. 13. Llobet E. Temperature-modulated semiconductor gas sensors. In: Grimes CA, Dickery EC, Pishko MV, editors. Encycl. Sensors. Los Angeles: American Scientific Pub- lishers; 2006. p. 131e52. 14. Engel M, Baumbach M, Kammerer T, Schutze€ A. Preparation of microstructured pellistors and their application for fast fuel vapor discrimination. In: 17th IEEE Int. Conf. Micro electro Mech. Syst. Maastricht MEMS 2004 Tech. Dig.; 2004. p. 268e71. https:// doi.org/10.1109/MEMS.2004.1290574. 15. Fricke T, Sauerwald T, Schutze€ A. Study of pulsed operating mode of a microstructured pellistor to optimize sensitivity and poisoning resistance. Proc IEEE Sensors 2014:661e4. https://doi.org/10.1109/ICSENS.2014.6985085. 16. Bastuck M, Bur C, Lloyd Spetz A, Andersson M, Schutze€ A. Gas identification based on bias induced hysteresis of a gas-sensitive SiC field effect transistor. J Sensor Sens Syst 2014;3:9e19. https://doi.org/10.5194/jsss-3-9-2014. 17. Bur C. Selectivity enhancement of gas sensitive field effect transistors by dynamic operation. Dissertation Linkoping€ University and Saarland University , Linkoping€ University Elec- tronic Press/Shaker Verlag; 2015. http://liu.diva-portal.org/smash/record.jsf? pid¼diva2:791934&dswid¼-3121. Dynamic operation of semiconductor sensors 409

18. Fischer S, Pohle R, Fleischer M, Moos R. Method for reliable detection of different exhaust gas components by pulsed discharge measurements using standard zirconia based sensors. Procedia Chem 2009;1:585e8. https://doi.org/10.1016/j.proche. 2009.07.146. 19. Schutze€ A. Pr€aparation und Charakterisierung von Phthalocyanin-Schichten zum Nachweis oxi- dierender und reduzierender Gase. Aachen: Dissertation, Justus-Liebig-Universit€at Gießen, Shaker Verlag; 1995. 20. Pohle R, von Sicard O, Fleischer M, Frerichs HP, Wilbertz C, Freund I. Gate pulsed readout of floating gate FET gas sensors. Procedia Eng 2010;5:13e6. https://doi.org/ 10.1016/j.proeng.2010.09.036. 21. Carotta MC, Cervi A, Fioravanti A, Gherardi S, Giberti A, Vendemiati B, Vincenzi D, Sacerdoti M. A novel ozone detection at room temperature through UV-LED-assisted ZnO thick film sensors. Thin Solid Films 2011;520:939e46. https://doi.org/10.1016/ j.tsf.2011.04.173. 22. Herran J, Fernandez-Gonzalez O, Castro-Hurtado I, Romero T, Mandayo GG, Castano~ E. Photoactivated solid-state gas sensor for carbon dioxide detection at room temperature. Sensor Actuator B Chem 2010;149:368e72. https://doi.org/10.1016/ j.snb.2010.06.050. 23. Vergara A, Benkstein KD, Semancik S. Thermally-assisted transient analysis for reducing the response time of microhotplate gas sensors. In: 2013 IEEE sensors; 2013. p. 1e4. https://doi.org/10.1109/ICSENS.2013.6688156. 24. Gutierrez-Osuna R, Gutierrez-Galvez A, Powar N. Transient response analysis for temperature-modulated chemoresistors. Sensor Actuator B Chem 2003;93:57e66. https://doi.org/10.1016/S0925-4005(03)00248-X. 25. Kato Y, Yoshikawa K, Kitora M. Temperature-dependent dynamic response enables the qualification and quantification of gases by a single sensor. Sensor Actuator B Chem 1997;40:33e7. https://doi.org/10.1016/S0925-4005(97)80196-7. 26. Nakata S, Ozaki E, Ojima N. Gas sensing based on the dynamic nonlinear responses of a semiconductor gas sensor: dependence on the range and frequency of a cyclic temperature change. Anal Chim Acta 1998;361:93e100. https://doi.org/10.1016/ S0003-2670(98)00013-0. 27. Kunt TA, McAvoy TJ, Cavicchi RE, Semancik S. Optimization of temperature programmed sensing for gas identification using micro-hotplate sensors. Sensor Actuator B Chem 1998;53:24e43. https://doi.org/10.1016/S0925-4005(98)00244-5. 28. Schweizer-Berberich M, Zdralek M, Weimar U, Gopel€ W, Viard T, Martinez D, Seube A, Peyre-Lavigne A. Pulsed mode of operation and artificial neural network evaluation for improving the CO selectivity of SnO2 gas sensors. Sensor Actuator B Chem 2000;65:91e3. https://doi.org/10.1016/S0925-4005(99)00333-0. 29. Fischer S, Pohle R, Magori E, Schonauer-Kamin€ D, Fleischer M, Moos R. Pulsed polarization of platinum electrodes on YSZ. Solid State Ionics 2012;225:371e5. https://doi.org/10.1016/j.ssi.2012.03.020. 30. Morrison SR. Semiconductor gas sensors. Sensor Actuator 1981;2:329e41. https:// doi.org/10.1016/0250-6874(81)80054-6. 31. Heiland G, Kohl D. Problems and possibilities of oxidic and organic semiconductor gas sensors. Sensor Actuator 1985;8:227e33. https://doi.org/10.1016/0250-6874(85) 85005-8. 32. Madou MJ, Morrison SR. Chemical sensing with solid state devices. San Diego: Academic Press; 1989. 33. Barsan N, Weimar U. Conduction model of metal oxide gas sensors. J Electroceram 2001; 7:143e67. https://doi.org/10.1023/A:1014405811371. 34. Yamazoe N, Shimanoe K. Theory of power laws for semiconductor gas sensors. Sensor Actuator B Chem 2008;128:566e73. https://doi.org/10.1016/j.snb.2007.07.036. 410 Andreas Schutze€ and Tilman Sauerwald

35. Heiland G. Homogeneous semiconducting gas sensors. Sensor Actuator 1981;2:343e61. https://doi.org/10.1016/0250-6874(81)80055-8. 36. Bârsan N, Hubner€ M, Weimar U. Conduction mechanisms in SnO2 based polycrystalline thick film gas sensors exposed to CO and H2 in different oxygen backgrounds. Sensor Actuator B Chem 2011;157:510e7. https://doi.org/10.1016/j.snb.2011.05.011. 37. Gramm A, Schutze€ A. High performance solvent vapor identification with a two sensor array using temperature cycling and pattern classification. Sensor Actuator B Chem 2003; 95:58e65. https://doi.org/10.1016/S0925-4005(03)00404-0. 38. Polese D, Martinelli E, Catini A, D’Amico A, Di Natale C. Self-adaptive thermal modulation of gas sensors. Procedia Eng 2010;5:156e9. https://doi.org/10.1016/j.pro eng.2010.09.071. 39. Paczkowski S, Paczkowska M, Dippel S, Schulze N, Schutz€ S, Sauerwald T, Weiß A, Bauer M, Gottschald J, Kohl C-D. The olfaction of a fire beetle leads to new concepts for early fire warning systems. Sensor Actuator B Chem 2013;183:273e82. https:// doi.org/10.1016/j.snb.2013.03.123. 40. Jaegle M, Wollenstein€ J, Meisinger T, Bottner€ H, Muller€ G, Becker T, Braunmuhl€ CB- V. Micromachined thin film SnO2 gas sensors in temperature-pulsed operation mode. Sensor Actuator B Chem 1999;57:130e4. https://doi.org/10.1016/S0925-4005(99) 00074-X. 41. Leidinger M, Sauerwald T, Reimringer W, Ventura G, Schutze€ A. Selective detection of hazardous VOCs for indoor air quality applications using a virtual gas sensor array. J Sensor Sens Syst 2014;3:253e63. https://doi.org/10.5194/jsss-3-253-2014. 42. Bastuck M, Leidinger M, Sauerwald T, Schutze€ A. Improved quantification of naphthalene using non-linear partial least squares regression. In: 16th Int. Symp. Olfaction Electron. Nose, Dijon, Fr.; 2015. http://arxiv.org/abs/1507.05834. 43. Sauerwald T, Baur T, Schutze€ A. Strategien zur Optimierung des temperaturzyklischen Betriebs von Halbleitergassensoren. In: XXVIII. Messtechnisches Symp. des Arbeitskreises der Hochschullehrer f€ur Messtechnik. Aachen: Shaker Verlag; 2014. p. 65e74. https:// doi.org/10.5162/AHMT2014/3.1. 44. Schultealbert C, Baur T, Schutze€ A, Bottcher€ S, Sauerwald T. A novel approach towards calibrated measurement of trace gases using metal oxide semiconductor sensors. Sensor Actuator B Chem 2017;239:390e6. https://doi.org/10.1016/ j.snb.2016.08.002. 45. Baur T, Schutze€ A, Sauerwald T. Detektion von kurzen Gaspulsen fur€ die Spurenga- sanalytik (detection of short gas pulses for trace gas analysis). Tm e Tech Mess 2017; 84(S1):88e92. https://doi.org/10.1515/teme-2017-0035. 46. Nakata S, Akakabe S, Nakasuji M, Yoshikawa K. Gas sensing based on a nonlinear response: discrimination between hydrocarbons and quantification of individual components in a gas mixture. Anal Chem 1996;68:2067e72. https://doi.org/ 10.1021/ac9510954. 47. Nakata S, Takemura K, Neya K. Non-linear dynamic responses of a semiconductor gas sensor: evaluation of kinetic parameters and competition effect on the sensor response. Sensor Actuator B Chem 2001;76:436e41. https://doi.org/10.1016/S0925-4005(01) 00652-9. 48. Ding J, McAvoy TJ, Cavicchi RE, Semancik S. Surface state trapping models for SnO2- based microhotplate sensors. Sensor Actuator B Chem 2001;77:597e613. https:// doi.org/10.1016/S0925-4005(01)00765-1. 49. Baur T, Schutze€ A, Sauerwald T. Optimierung des temperaturzyklischen Betriebs von Halbleitergassensoren (optimization of temperature cycled operation of semiconductor gas sensors). Tm e Tech Mess 2015;82:187e95. https://doi.org/10.1515/teme-2014- 0007. Dynamic operation of semiconductor sensors 411

50. Schultealbert C, Baur T, Schutze€ A, Sauerwald T. Facile quantification and identification techniques for reducing gases over a wide concentration range using a MOS sensor in temperature-cycled operation. Sensors (Switzerland) 2018;18:744. https://doi.org/ 10.3390/s18030744. 51. Sauerwald T, Baur T, Leidinger M, Reimringer W, Spinelle L, Gerboles M, Kok G, Schutze€ A. Highly sensitive benzene detection with metal oxide semiconductor gas sensors e an inter-laboratory comparison. J Sensor Sens Syst 2018;7:235e43. https:// doi.org/10.5194/jsss-7-235-2018. 52. Baur T, Schultealbert C, Schutze€ A, Sauerwald T. Novel method for the detection of short trace gas pulses with metal oxide semiconductor gas sensors. J Sensor Sens Syst 2018;7:411e9. https://doi.org/10.5194/jsss-7-411-2018. 53. Leidinger M, Rieger M, Sauerwald T, Alépée C, Schutze€ A. Integrated pre-concentrator gas sensor microsystem for ppb level benzene detection. Sensor Actuator B Chem 2016;236:988e96. https://doi.org/10.1016/j.snb.2016.04.064. 54. Bogner€ M, Doll T. Advanced gas sensing e introduction to the electroadsorptive effect and its application. In: Doll T, editor. Adv. Gas Sens. e electroadsorptive Eff. Relat. Tech. Springer; 2003. p. 2e37. 55. Liess M. Electric-field-induced migration of chemisorbed gas molecules on a sensitive filmda new chemical sensor. Thin Solid Films 2002;410:183e7. https://doi.org/ 10.1016/S0040-6090(02)00209-2. 56. Sauerwald T, Skiera D, Kohl D. Field induced polarisation and relaxation of tungsten oxide thick films. Thin Solid Films 2005;490:86e93. https://doi.org/10.1016/ j.tsf.2005.04.009. 57. Sauerwald T, Skiera D, Kohl CD. Selectivity enhancement of gas sensors using nonequilibrium polarisation effects in metal oxide films. Appl Phys A Mater Sci Process 2007;87:525e9. https://doi.org/10.1007/s00339-007-3980-2. 58. Fort A, Gregorkiewitz M, Machetti N, Rocchi S, Serrano B, Tondi L, Ulivieri N, Vignoli V, Faglia G, Comini E. Selectivity enhancement of SnO2 sensors by means of operating temperature modulation. Thin Solid Films 2002;418:2e8. https:// doi.org/10.1016/S0040-6090(02)00575-8. 59. Zhang S, Lei T, Li D, Zhang G, Xie C. UV light activation of TiO2 for sensing formaldehyde: how to be sensitive, recovering fast, and humidity less sensitive. Sensor Actuator B Chem 2014;202:964e70. https://doi.org/10.1016/j.snb.2014.06.063. 60. Klaus D, Klawinski D, Amrehn S, Tiemann M, Wagner T. Light-activated resistive ozone sensing at room temperature utilizing nanoporous In2O3 particles: influence of particle size. Sensor Actuator B Chem 2015;217:181e5. https://doi.org/10.1016/ j.snb.2014.09.021. 61. Klawinski D, Meixner D, Kohl C, Wagner T. Cyclic optical activation of semiconducting gas sensors: influence of cycling frequency. In: Proc. Sensors 2017, AMA Conf. 2017- 05-30 e 2017-06-01 Nuremberg, ger; 2017. p. 731e3. https://doi.org/10.5162/ sensor2017/P5.11. 62. Gonzalez O, Roso S, Vilanova X, Llobet E. Enhanced detection of nitrogen dioxide via combined heating and pulsed UV operation of indium oxide nano-octahedra. Beilstein J Nanotechnol 2016;7:1507e18. https://doi.org/10.3762/bjnano.7.144. 63. Chinh ND, Quang ND, Lee H, Thi Hien T, Hieu NM, Kim D, Kim C, Kim D. NO gas sensing kinetics at room temperature under UV light irradiation of In2O3 nanostructures. Sci Rep 2016;6:35066. https://doi.org/10.1038/srep35066. 64. Schuler€ M, Sauerwald T, Schutze€ A, Gaudillat P, Suisse J-M, Bouvet M. Selective quantification of humidity and ammonia by optical excitation of molecular semiconductor-doped insulator (MSDI) sensors. 2015 IEEE sensors e Proc. 2015: p1e4. https://doi.org/10.1109/ICSENS.2015.7370407. 412 Andreas Schutze€ and Tilman Sauerwald

65. Lundstrom€ I, Shivaraman S, Svensson C. A hydrogen-sensitive Pd-gate MOS transistor. J Appl Phys 1975;46:3876e81. https://doi.org/10.1063/1.322185. 66. Eklov€ T, Mårtensson P, Lundstrom€ I. Enhanced selectivity of MOSFET gas sensors by systematical analysis of transient parameters. Anal Chim Acta 1997;353:291e300. https://doi.org/10.1016/S0003-2670(97)87788-4. 67. Briand D, Wingbrant H, Sundgren H, Van der Schoot B, Ekedahl LG, Lundstrom€ I, De Rooij NF. Modulated operating temperature for MOSFET gas sensors: hydrogen recovery time reduction and gas discrimination. Sensor Actuator B Chem 2003;93: 276e85. https://doi.org/10.1016/S0925-4005(03)00230-2. 68. Bur C. New method for selectivity enhancement of SiC field effect gas sensors for quantification of NOx. Diploma thesis. Saarland University; 2012. 69. Bur C, Reimann P, Andersson M, Schutze€ A, Lloyd Spetz A. Increasing the selectivity of Pt-gate SiC field effect gas sensors by dynamic temperature modulation. IEEE Sens J 2012:1267e72. https://doi.org/10.1109/JSEN.2011.2179645. 70. Bur C, Bastuck M, Puglisi D, Schutze€ A, Lloyd Spetz A, Andersson M. Discrimination and quantification of volatile organic compounds in the ppb-range with gas sensitive SiC-FETs using multivariate statistics. Sensor Actuator B Chem 2015;214:225e33. https://doi.org/10.1016/j.snb.2015.03.016. 71. Lloyd Spetz A, Huotari J, Bur C, Bjorklund R, Lappalainen J, Jantunen H, Schutze€ A, Andersson M. Chemical sensor systems for emission control from combustions. Sensor Actuator B Chem 2013;187:184e90. https://doi.org/10.1016/j.snb.2012.10.078. 72. Darmastuti Z, Bur C, Moller€ P, Rahlin R, Lindqvist N, Andersson M, Schutze€ A, Lloyd Spetz A. SiC-FET based SO2 sensor for power plant emission applications. Sensor Actu- ator B Chem 2014;194:511e20. https://doi.org/10.1016/j.snb.2013. 11.089. 73. Eriksson M, Petersson LG. Spillover of hydrogen, oxygen and carbon monoxide in oxidation reactions on SiO2 supported Pd. Surf Sci 1994;311:139e52. https:// doi.org/10.1016/0039-6028(94)90485-5. 74. Eriksson M, Salomonsson A, Lundstrom€ I, Briand D, Åbom AE. The influence of the insulator surface properties on the hydrogen response of field-effect gas sensors. J Appl Phys 2005;98:1e7. https://doi.org/10.1063/1.1994941. 75. Daut C. Signalkompensation mittels Gate-bias bei gassensitiven SiC-Feldeffekttransistoren. Bachelor thesis. Saarland University; 2017. 76. Bastuck M, Daut C, Schutze€ A. Signalkompensation mittels gate-potential bei gassensitiven Feldeffekttransistoren. In: 13. Dresdner sensor-symposium; 2017. p. 277e82. https://doi.org/10.5162/13dss2017/P4.03. 77. Bur C, Bastuck M, Lloyd Spetz A, Andersson M, Schutze€ A. Selectivity enhancement of SiC-FET gas sensors by combining temperature and gate bias cycled operation using multivariate statistics. Sensor Actuator B Chem 2014;193:931e40. https://doi.org/ 10.1016/j.snb.2013.12.030. 78. Kammerer T, Engel M, Schutze€ A. An intelligent fuel sensor based on a microstructured gas sensor. In: Proc. IEEE sensors 2003 (IEEE Cat. No.03CH37498), vol. 2; 2003. p. 1064e9. https://doi.org/10.1109/ICSENS.2003.1279106. 79. Schuler€ M, Sauerwald T, Schutze€ A. Metal oxide semiconductor gas sensor self-test using fourier-based impedance spectroscopy. J Sensor Sens Syst 2014;3:213e21. https:// doi.org/10.5194/jsss-3-213-2014. 80. Leidinger M, Reimringer W, Alépée C, Rieger M, Sauerwald T, Conrad T, Schutze€ A. Gas measurement system for indoor air quality monitoring using an integrated preconcentrator gas sensor system. In: Conf. Mikro-Nano-Integration e 6. GMM-workshop 10/05/2016 e 10/06/2016 Duisburg, Deutschl; 2016. p1e6. In: https://www.vde- verlag.de/proceedings-en/454278027.html. CHAPTER THIRTEEN

Micromachined semiconductor gas sensors

D. Briand1, J. Courbat2 1Ecole Polytechnique Fédérale de Lausanne, Neuch^atel, Switzerland 2Formely Ecole Polytechnique Fédérale de Lausanne, Neuch^atel, Switzerland

Contents

13.1 Introduction 413 13.2 A brief history of semiconductors as gas-sensitive devices 414 13.3 Microhotplate concept and technologies 416 13.3.1 Concept and thermal design 416 13.3.2 Microhotplate realization and performance 418 13.3.3 Microhotplate reliability 421 13.4 Micromachined metal oxide gas sensors 425 13.4.1 Thin gas-sensitive films 425 13.4.2 Thick gas-sensitive films 428 13.4.3 Temperature modulation 432 13.4.4 Packaging 435 13.5 Complementary metal oxide semiconductorecompatible metal oxide gas 437 sensors 13.6 Micromachined field-effect gas sensors 442 13.7 Nanostructured gas sensing layers on microhotplates 445 13.8 Semiconductor gas sensors on polymeric foil and their additive manufacturing 450 13.8.1 Semiconductor gas sensors on polymeric foil 450 13.8.2 Printing semiconductor gas sensors 452 13.9 Manufacturing, products, and applications 454 13.10 Conclusion 458 References 459

13.1 Introduction Metal oxide gas sensors based on screen printing thick layers on alumina substrates to form a platinum heater and electrodes, and to pattern the thick metal oxide gasesensitive ﬁlm, have been commercialized for a few decades. At the beginning of the 1980s, micromachining of silicon

Semiconductor Gas Sensors, Second Edition ISBN: 978-0-08-102559-8 © 2020 Elsevier Ltd. https://doi.org/10.1016/B978-0-08-102559-8.00013-6 All rights reserved. 413 j 414 D. Briand and J. Courbat took considerable strides and led to the emergence of new microelectromechanical systems (MEMS) devices. The use of microfabrication techniques to realize microsensors and MEMS devices has brought different advantages than miniaturization, such as batch processing, formation of arrays, reduced power consumption, and new modes of operation. Some work has been undertaken by micromachining anodic alumina1,2 but the extensive developments were carried out based on silicon micromachining.3 This chapter therefore focuses on silicon micromachined semiconductor gas sensors. After a brief history of silicon hotplates and metal oxide gas sensors, more information will be provided on the microhotplate concept, realization, and reliability. The core of this chapter comprises a section on micromachined thin- and thick-ﬁlm metal oxide gas sensors addressing temperature modulation. Some highlights are given concerning complementary metal oxide semiconductor (CMOS) and silicon on insulator (SOI) implementation of metal oxide gas sensors and micromachined ﬁeld-effect gas sensors. Finally, trends on the integration of nanostructured gas sensing materials on micromachined transducers and on semiconductor gas sensors on polymeric foil, and their additive fabrication, are highlighted.

13.2 A brief history of semiconductors as gas-sensitive devices In 1952, Brattain and Bardeen reported on the change of the semiconducting properties of germanium with a variation of the partial pressure of oxygen in the surrounding atmosphere.4 Seiyama published 10 years later results demonstrating the gas sensing effect on metal oxides.5 Taguchi brought metal oxide semiconductor gas sensors to market using an alumina ceramic tube mounted with the metal oxide and electrodes and a heater coil passing through it. He founded in 1969 the company Figaro Engineering Inc., which is still today the largest manufacturer of semiconductor gas sensors worldwide. Nowadays, the commercially available devices are mostly manufactured using screen printing on small and thin ceramic substrates exhibiting a power consumption of 0.2e1 W. In 1988, Demarne et al. demonstrated and patented the first thin-film metal oxide gas sensors based on a micromachined silicon substrate. The microhotplate was made of a thermally insulating silicon oxide membrane. It embedded a gold heater. Gold electrodes were patterned on top and covered with a thin tin dioxide film. The device operated with a significantly reduced power consumption of about 100 mW to reach 300 C, a value still much lower than commercially available devices on alumina substrates. Micromachined semiconductor gas sensors 415

10 mm Mesh SnO2 layer Nylon cap charcoal Filter

15 mm Mesh Metal can Pt electrode Gold wire Metal Sensor die header

Poly-Si heater Bulk Si/SiO2 Si/SiO2 diaphragm 1 mm Figure 13.1 Diagram of the MGS 1100 sensor from Motorola. Micromachined sensor element is illustrated on the left, and the sensor housing on the right. The sensitive ﬁlms were obtained by rheotaxial growth and thermal oxidation of tin layers deposited on the silicon oxideenitride membrane. From Simon I, Barsan N, Bauer M, Weimar, U. Micromachined metal oxide gas sensors: opportunities to improve sensor performance. Sens Actuators B 2001;73:1e26.

Motorola licensed the technology and put effort into developing mass produced metal oxide gas sensors using silicon micromachining (Fig. 13.1). Polysilicon heaters were introduced in an oxideenitride membrane, using gold electrodes as before. They ceased work on the chemical sensor in 1998, but the technology was taken over by MicroChemical Systems SA in Switzerland and has evolved to be aligned with the developments reported by other research and industrial groups. Micromachined thick-film semiconductor gas sensors were introduced by drop-coating the metal oxide on a thin dielectric membrane with platinum used both for heaters and electrodes, offering improved performances and robustness.6 This technology has been exploited because then by AppliedSensor GmbH (Section 6.4.2). Temperature modulation was introduced as a mode of operation due to the low thermal mass of the microhotplates. This mode of operation is now mainly applied to applicative scenarios to minimize power consumption; to reduce the influence of humidity, for example, to enhance the discrimination capabilities of these sensors; and to improve their stability over time (Section 6.4.3). Since 2000, the field has been evolving toward the use of SOI wafers, the implementation of these sensors in CMOS technology and on polymeric substrates, and the identification of suitable modes of operation for different applications. The field is now strongly focusing on nanomaterials,7 especially 416 D. Briand and J. Courbat nanostructured metal oxides, but one can question whether this would be the solution to the main problems remaining with thin- and thick-film devices. Despite the extensive work carried out in this regard, little has transferred to and been exploited by industry so far. However, since 2010, different companies have been gaining interest in micromachined semiconductor gas sensors, such as AMS in Austria, Bosch in Germany, Figaro in Japan, and Sensirion AG in Switzerland. Microhotplates being a mature and robust technology, the main issue remains of the synthesis of performing materials and their effective integration into a robust manufacturing process. One trendy approach is the use of digital printing, i.e., inkjet, to deposit metal oxide nanoparticles in solution. Research and developments since the end of the 1980s has reported a huge set of metal oxide materials and hotplate combinations. Because of limitations of space, it has been necessary to be selective regarding the work to be presented in this chapter, which is far from exhaustive. More details on the different configurations of alumina- and silicon-type metal oxide gas sensors can be found in Ref. 3.

13.3 Microhotplate concept and technologies Silicon micromachining has been used to generate thermally insulated heating elements suspended on a dielectric membrane. By patterning metallic electrodes (Au, Pt) on top of the membrane, these structures have been applied as low-power transducers in metal oxide gas sensors. This section provides information on the design, fabrication, characteristics, and reliability of microhotplates used in semiconductor gas sensors. 13.3.1 Concept and thermal design The operation of a metal oxide gas sensor relies on the change in resistance of an n-orp-type semiconducting layerdmainly SnO2dwhen exposed to reducing or oxidizing gases. A diagram of a typical cross-sectional view of a silicon micromachined metal oxide (MOX) sensor is presented in Fig. 13.2. Their development has evolved toward silicon substrates to produce devices suitable for commercialization due to their low cost, low-power consumption, and high reliability. To lower the resistivity of the gas-sensitive ﬁlm, as well as to improve the kinetics of the chemical reactions, the metal oxide layer is heated with a microheater. The heated area is usually embedded in a thin dielectric membrane to improve the thermal insulation and to reduce the power Micromachined semiconductor gas sensors 417

Gas sensitive layer Dielectric Electrodes membrane Heater

Figure 13.2 Cross-sectional diagram of a micromachined metal oxide gas sensor.

Convection Radiation

Thot

Conduction

Tamb

Figure 13.3 Heat losses in a microheating device: conduction, convection, and radiation. consumption of the device, which is typically in the order of a few tens of milliwatts at 300 C, and its thermal time constant (few to tens of milliseconds). Thermal programming allows kinetically controlled selectivity. Fig. 13.3 illustrates the heat losses that occur in a microhotplate when operating. The thermal energy, Q, generated by the Joule effect in the microheater, is given by DQ ¼ R$I2$Dt (13.1) where I is the current ﬂowing through the heater with a resistance R during Dt time. This heat is dissipated in the device and in the surrounding environment by three means: • conduction in the device; • convection in the surrounding media (typically air); and • radiation. 418 D. Briand and J. Courbat

Thus, the heat generated by the microheater is equal to the sum of the heat lost by conduction in the device, Qcond, by convection in the air, Qconv, and by radiation, Qrad: 2 R $ I $Dt ¼ DQcond þ DQconv þ DQrad (13.2) The thermal design of microhotplates is mainly based on finite element simulation with the objective of optimizing the power consumption and obtaining a uniform temperature distribution over the active area. A precise model to evaluate the uniformity of power consumption and temperature over the heated area requires many empirical parameters to be known or measured accurately.8,9 Different heater layouts have been published, mainly meander or spiral shapes6,10 spiral shapes exhibiting better spatial temperature uniformity.11,12 Improvement in temperature uniformity was also attempted by using a plate heater as shown by Ç akir et al.13. A maximum temperature variation of 7% was reached in the sensor-active area using an ITO-based heater. Also the implementation of an array of sensors on a single membrane/heater has been considered to decrease size/cost and overall power consumption. 13.3.2 Microhotplate realization and performance Microhotplates are made using a combination of thin-film and silicon micromachining processes. There are two main kinds of micromachined silicon substrates: closed membrane and bridge membrane. They consist of a suspended thin dielectric membrane, made of silicon nitride and/or silicon oxide, that is released using silicon micromachining on either the obverse or reverse faces. The typical thickness of the membranes is from 0.5 to 2 mm. Closed membranes have lateral dimensions of about 0.5e1 mm, with approximately half the length being used as the active area. Edge effects can be minimized by using circular membranes.14 The typical lateral dimensions of bridge membranes lie between 100 and 200 mm. A silicon plug/island or a highly thermal conductive material, such as silicon carbide, can be implemented to improve uniformity of temperature. Diagrams of these structures are presented in Fig. 13.4.A bridge membrane exhibits lower power consumption due to better thermal insulation from the silicon substrate, whereas a closed membrane is more convenient for patterning the sensing element. In addition, silicon micro- electronics components can be integrated on the thermally insulated area of the device. Amor et al.15 integrated temperature-measurement diodes and metal oxideesemiconductor field-effect transistor (MOSFET) under Micromachined semiconductor gas sensors 419

(a) (b)

Sensing material (thin or thick film) Electrodes Heater + thermometer

~ 1–2 μm Active area ~ 400 μm Si ~ 1–1.5 mm

Membrane ~ 1–1.5 mm

~ 1–2 μm

Si ~ 400 μm Si

Si plug

Sensing material (thin or thick film) Electrodes Heater + thermometer

Suspension beams Active area Pit

Anisotropic etching

~ 100–200 μm

Si Si Sacrificial etching Figure 13.4 Diagram of a suspended membraneetype gas sensor; (a and b) reverse of silicon micromachining; (c and d) obverse surface micromachiningd(a and c) top view, (b and d) side view. Adapted from Simon I, Barsan N, Bauer M, Weimar, U. Micromachined metal oxide gas sensors: opportunities to improve sensor performance. Sens Actuators B 2001;73:1e26. their microheater that could be heated up to 335 C. N-MOSFET and p-MOSFET showed good properties up to 280 and 240 C, respectively. Microhotplates with a bridge-membrane design based on CMOS- compatible processes were proposed by Cavicchi et al.16. The architecture of the hotplate is presented in Fig. 13.5. During the 2000s, the Swiss Federal Institute of Technology Zurich (ETHZ), Switzerland, came up with different generations of CMOS micromachined metal oxide gas sensors 420 D. Briand and J. Courbat

(a)

Suspended structure

50 μm (b)

SnO2 oxide film

Film contacts

Insulating SiO2 Doped polysilicon heater

Insulating SiO2 Figure 13.5 Obverse of CMOS silicon micromachined hotplate: (a) optical picture; (b) diagram. Courtesy of Dr Steve Semancik, NIST, USA. with integrated driving and readout circuitries.17 The heat necessary for the chemical reactions between the gaseous environment and the sensing layer was provided by the Joule effect through a field-effect transistor (FET) or polysilicon resistor. For improved reliability, platinum and tungsten are preferred as heater material at the time of writing. More details on the heater performances are provided in Section 6.3.3, on reliability. The heater and thermometer, which are needed to control the sensor operation temperature, can be either a dual purpose unit or two separate components. Poly- silicon and platinum have often been used; microelectronic components, such as a forward bias silicon pen junction as a temperature sensor, can be considered when silicon is available on the membrane. With a resolution in the micrometer range for the photolithographic patterning of the electrodes, the gas-sensitive area can be significantly reduced in comparison with screen printing on ceramic substrates. Regarding the electrode material, platinum is favored because it shows very good chemical stability and can provide higher gas responses.18 The two main approaches for the deposition of the gas-sensitive sensing layer Micromachined semiconductor gas sensors 421 are either thin- or thick-film techniques. A thin film is usually realized by evaporation or sputtering; a thick film is deposited by screen printing, spray pyrolysis, or drop coating.3 Once deposited, these materials usually require annealing at high temperatures (350e800 C) in an oxygen-containing atmosphere to modify the morphology (e.g., grain size) and microstructure (e.g., porosity, surface-to-volume ratio). The parameters of this annealing step have to be carefully selected to be compatible with the microhotplate itself. Some temperature limitations occur with microhotplates based on a CMOS-compatible process. Several micromachined hotplates for metal oxide gas sensors have been reported in the literature. However, robust and established technologies all make use of the closed-membrane design in combination with platinum as the electrode material. Recent papers show that platinum is now mainly used as a heater material with tungsten applied in CMOS-compatible devices. The characteristics of some representative examples are summarized in Table 13.1. The optimization of the micromachined platform is very close to the optimum achievable, with a minimum active areadand, therefore, power consumptiondreached. According to the resolution of the photolithographic process, it is becoming difficult to further reduce the size of the hotplates and yet retain an exploitable sensing layer and heater resistance values. The next steps are toward using nanopatterning techniques, self-heated metal oxide nanostructures, and printing on flexible polymeric substrates, as presented in Sections 6.7 and 6.8. 13.3.3 Microhotplate reliability Operating at a relatively high temperature, the electrothermomechanical reliability of micromachined hotplates is an important aspect for metal oxide gas sensors. Numerical thermomechanical studies have been performed to improve the robustness of the membrane, addressing buckling and stress concentration.19 Thermomechanical reliability depends on the design and materials used. In general, the membranes made of dielectric materials deposited at a higher temperature (e.g., low-pressure chemical vapor depositiondLPCVD) are more robust. Attempts with SiN deposition were also performed by PECVD, however, they revealed to be less robust.20 The membrane is usually formed of a stress-compensated stack of thin films of silicon nitride, silicon oxynitride, and/or silicon oxide. A heater embedded in between LPCVD low-stress silicon nitride thin films has proven to be robust.6,21 This dielectric material is, however, not commonly available in MEMS foundries. Different Table 13.1 Comparison of various microhotplate designs that have been reported in the literature. 422 Power/heater Material Active area Hotplate area Power at area (mW/ CMOS Membrane membrane or Year Author (1000 mm2) (1000 mm2) 300 C (mW) 1000 mm2) Yes/No (M)/bridge (B) bridge Heater

1990 Dibbern 202.5 1822.5 55 0.27 No M Oxynitride NiFe 1993 Suehle 10 40 40 4.00 Yes B CMOS films Poly-Si 1995 Zanini 722.5 1440 90 0.12 No M Oxynitride Pt 1995 Gardner 472.5 3596.4 40 0.08 No M Nitride Pt 1996 Aigner 300 1000 35 0.12 No M Nitride Pt 1996 Lee 10 1000 18 1.80 No M Oxynitride Poly-Si Pt 1997 Gotz 250 1210 55 0.22 No M Oxynitride Poly-Si 1998 Guidi 562.5 2250 67 0.12 No M Nitride Pt 1998 Astie 230 3240 125 0.54 No M Si/SiO Poly-Si 1999 Horrillo 250 1210 38 0.15 No M Nitride Poly-Si 2001 Udrea 90 250 100 1.11 Yes M CMOS films FET 2001 Benn 40 NA 8.6 0.215 No B SiC SiCeN 2002 Afridi 10 44 27.5 2.75 Yes B CMOS films Poly-Si 2002 Briand 202.5 1000 50 0.25 No M Nitride Pt/FET 2002 Mo 6.4 25.6 6 0.94 No B Oxynitride Pt 2002 Chan 14.4 57.6 60 4.17 No B Oxynitride Poly-Si 2003 Lee 31.4 1000 30 0.95 No M Nitride Pt 2003 Tsamis 10 NA 15 1.50 No B Porous Si Poly-Si Pt Courbat J. and Briand D. 2004 Fujres 10 NA 7.5 0.75 No B Nitride Pt 2004 Baroncini 250 1000 20 0.08 No M Nitride Pt 2004 Laconte 57.6 409.6 13 0.22 No M Oxynitride Poly-Si 2005 Graf 70.7 250 50 0.71 Yes M CMOS films Poly-Si 2005 Lee 3990 NA 73 0.02 No B Nitride Pt irmcie eiodco a sensors gas semiconductor Micromachined 2006 Elmi 20.1 NA 6 0.30 No B Oxynitride Pt 2006 Belmonte 160 NA 30 0.19 No B Oxynitride Pt 2007 Guo 36.1 90 23 0.64 No B Oxynitride Pt 2008 Barborini 1000 NA 24 0.02 No B Oxynitride Pt 2008 Briand 250 2250 60 0.24 No M Polyimide Pt 2008 Ali 17.67 250 14 0.79 Yes M CMOS films W 2008 Ali 0.452 70.7 6 13.27 Yes M CMOS films W

Notes: Where exact values are not given, they have been deduced from the information given in the particular paper. NA: not available. Adapted from Ali SZ, Udrea F, Milne WI, Gardner JW. Tungsten based SOI microhotplates for smart gas sensors. J Microelectromech S 2008;17(6):1408e1417; in which all references can be found. 423 424 D. Briand and J. Courbat techniques have been implemented to improve mechanical stability of the membrane. Iwata et al.22 added SU-8 structures to reinforce bridges of a suspended membrane at a cost of a higher power consumption. Accelerated aging tests have also been developed to determine and analyze the failure mechanisms by thermally cycling the device, by ramping up the power until breakdown, or by operating it at temperatures higher e than their normal use.23 25 Cracks in the dielectric membrane, electromigration, and electroestress migration have been identified as the main causes of failure.26 At high temperatures, the migration of the platinum atoms in the heater meander was linked to the mechanical stress in the dielectric membrane. They usually occur in location of high temperature gradient and/or high current densities. Reduction of current density accumulation between two different conductive materials has been achieved by27. Platinum was used for the heater, while conductive tracks were made of AlCu. Current density at the metal junction could be reduced by 20% by forming a slope of 45 at the end of the AlCu line, reducing failure likeli- hood of the electrical connection. State-of-the-art technologies can allow temperature cycling up to several millions of cycles before failure, enabling temperature modulation of the sensor (Section 6.4.3). The heater material is a crucial point for the stability of this type of device during operation. Driven by CMOS compatibility, poly-Si was first used but it suffers from an inappropriate drift of its electrical resistivity at high temperature.28 Platinum is the material that has been implemented for the heater for improved reliability. It is used in most micromachined metal oxide sensors on the market at the time of writing, not only for the heater but also for the electrodes. Courbat et al.29 showed that adding a small amount of another refractory metal (such as iridium) to the platinum can improve its resistance to electromigration. However, Mo exhibited superior performances to platinum, allowing higher operational temperatures30 and low heater resistance drift.20 TiNda CMOS-compatible materialdhas been applied as a heating element showing relatively better performances than platinum.31 FETs have also been implemented as heaters in CMOS technology but this requires a silicon area in or underneath the membrane.32 A very low-power micromachined hotplate platform was designed using SOI technology and a robust tungsten heater.8 This device is on the market in the products portfolio from Cambridge CMOS Sensors Ltd in the United Kingdom, which was acquired in 2016 by AMS, Austria. One constraint is the obligation to work with the thin films available in the CMOS process. Depending on the Micromachined semiconductor gas sensors 425 process, the CMOS dielectric stack of films is not always optimum and post- processes can be necessary. For instance, this can involve the deposition of the metallic electrodes (Pt, Au), or a passivation and stress compensating dielectric thin film.

13.4 Micromachined metal oxide gas sensors In the main, two types of metal oxide gas-sensitive films have been integrated into micromachined hotplate transducers: thin and thick films. The different developments will be presented in this section. The integration of a third type of structurednanowires, into which considerable efforts are being made at the time of writingdwill be presented in Section 6.7.1, Trends and perspectives. Other chapters in this book address in detail the synthesis, sensing mechanisms, and properties related to these different sensing films. In this section, for better readability and to allow comparison between results, all responses are given as Rgas/Rair if Rgas > Rair or as Rair/Rgas if Rair > Rgas, where Rair is the baseline resistance of the sensors in air and Rgas is its resistance when exposed to the analyte under examination. 13.4.1 Thin gas-sensitive films First, micromachined gas sensors were obtained using thin-film deposition technologies. That technique, used for semiconductor manufacturing, is available in most cleanrooms with evaporation or sputtering machines. The motivation at that time was to produce MEMS-based metal oxide gas sensors using thin-film technology only, being a disruptive technology compared with the thick-film technologies used on alumina. The first silicon micromachined thin-film metal oxide gas sensor was developed at CSEM SA, Switzerland by Demarne et al.21; this was commercialized at the beginning of the 1990s by Microsens SA in Switzerland. It consisted in a SiO2 membrane embedding a gold-based meander-shaped heater. A thin film of SnOx was sputtered and patterned by lift-off. Two configurations were proposed, without and with a silicon plug to make the temperature reached in the active area of the device more uniform. To attain 300 C, the supplied powers were, respectively, 104 and 183 mW. Motorola also showed a significant interest in the development of commercial micromachined thin-film gas sensors for CO detection.33 They ceased their activities in that field at the end of the 1990s. Develop- ment was pursued by MiCS (MicroChemical Systems SA) in Switzerland, now part of the SGX Sensortech group. 426 D. Briand and J. Courbat

Other techniques have been used for the fabrication of thin-film metal oxide gas sensors. At NIST in the United States,16,34 produced gas sensors by chemical vapor deposition (CVD). By applying a current and thus heating the hotplate, sensing films could be deposited locally (i.e., only on heated active areas) using an adequate organometallic precursor. SnO2 and ZnO films were obtained with tetramethyltin and diethylzinc in an oxygen atmosphere. They were deposited onto different seed layers, which played a significant role in terms of gas selectivity. Besides CVD and sputtering from a target of the desired material, thin films were obtained by sputtering or evaporation through the rheotaxial growth and thermal oxidation (RGTO) process. This method consists in depositing thin layers of a metal, followed by its thermal oxidation in an oxygen-rich atmosphere. Tin oxide layers of 350 nm in thickness were obtained with this technique from sputtered Sn by Faglia et al.11 for the design of CO sensors. The highest sensitivity to CO was obtained at an operating temperature of 400 C. Responses of between two and three were obtained when the device was exposed to 25 ppm of CO, the alarm 35 level in many countries. With the same technique, grew SnO2 films on very low power hotplates. A temperature of 300 C was reached with a supply power of 6 mW. The sensor had a response of 7.5 when exposed to 100 ppb of NO2 at 200 C and 5 under 10 ppm of CO at 450 C. In 2003, the European Aeronautic Defence and Space company in Ger- many developed gas sensors based on silicon technology to replace thick- film devices, which were usually based on alumina substrates and had a high level of power consumption.36 A main drawback Muller et al. identified in Si-based devices was their fragile membrane. Therefore, they built their devicesdan array of three hotplatesdfrom SOI to keep the top Si layer as a robust suspended membrane. A fabrication yield of 100% was achieved with a top Si layer thicker than 5 mm. Typical power consumptions were in the range of 50 e80 mW to reach an operating temperature of 300 C. The active area of the device could be operated at different temperatures and functionalized through thin- and/or thick-film technology. 37 Friedberger et al. evaporated Sn and obtained SnO2 by RGTO. The sensing film had good sensitivity toward hydrocarbon and hydrogen, but a very low response to CO. Wollenstein€ et al.38 developed an array combining several gas sensing layers by successive photolithography steps and sputtering or e-beam evaporation. A device with four different metal oxide layers could be produced. The films had to be deposited in a specific order, depending on the Micromachined semiconductor gas sensors 427 temperature required for stabilization. The layer with the highest annealing temperature was deposited first. Titanium-doped chromium oxide was produced by successively evaporating Cr and Ti layers, which were subsequently annealed at 850 C. ZnO films were obtained by direct current (DC) magnetron sputtering from a Zn target combined with an Ar/O2 plasma. Pt-doped SnO2 films were obtained by radio frequency magnetron reactive sputtering from a SnO2 target followed by the deposition of a few tens of a nanometer of Pt. The sintering of ZnO and SnO2 films occurred at a temperature of 700 C and could be performed simultaneously. As for ZnO, WO3 was sputtered from a W target in an Ar/O2 plasma with a low deposition rate to ensure proper oxidation of the material. The last material that could be deposited was V2O5. It was performed by e-beam evaporation of vanadium under controlled oxygen pressure. To reach a fully oxidized film, the evaporation was followed by an additional oxidation treatment at 500 C in synthetic air. The silicon wafer was then bonded to a micromachined glass component acting as a structural element. To reduce power consumption as much as possible, the reverse of the Si wafer was wet etched in a KOH solution. Etching stopped at the dielectric thin films and at the highly doped Si layer. Gas sensing measurements are presented in Fig. 13.6. The sensors were operated at about 200 mW to reach

100k Ω) 10k

600 500 400 WO3 300 CTO Sensor response ( ZnO 200 V O H CO NO NH 2 5 2 2 3 SnO 100 ppm 50 ppm 2 ppm 100 ppm 2 100 10 15 20 25 30 35 Time (hours) Figure 13.6 Gas measurement obtained from the sensor array. The sensing materials exhibited different behavior toward the analytes. From Wollenstein€ J, Plaza JA, Cané C, Min Y, Bottner€ H, Tuller HL. A novel single chip thin ﬁlm metal oxide array. Sens Actuators B 2003;93:350e355 428 D. Briand and J. Courbat

a temperature of 400 C. They were exposed to H2, CO, NO2, and NH3 as testing gases. Discrimination can be made between them because some material resistive variation was observed only for specific gases. ZnO was the only layer exhibiting a response to NO2 and V2O5 to NH3. In the mid 2010s, there has been a renewed interest in using pulsed laser deposition to produce metal oxide films with various morphologies on micromachined silicon hotplates.39 SGX SensorTech SA in Switzerland and Bosch in Germany have notably evaluated this technique to manufacture thin film metal oxide gas sensors integrated on MEMS hotplates. 13.4.2 Thick gas-sensitive films In the mid-1990s, thick filmebased metal oxide sensors began to attract attention. There were issues regarding the stability and reproducibility of metal oxide thin films. New deposition methods brought from outside the semiconductor industry were useddmainly pipetting, drop coating, and screen printing. The first combination of a thick-film sensing layer combined with a microhotplate was carried out by Barsan (see 40), by pipetting pure SnO2, 0.2% Pt-doped SnO2, or 0.2% doped SnO2 on gold electrodes patterned on micromachined hotplates. A polycrystalline structure was obtained by sintering the SnO2 layers at 600 C in air. A power supply of 60 mW was needed to operate the sensor at 400 C. The pure SnO2-based sensors showed the best sensitivity to organic solvents. It exhibited a resistance variation of 32% when exposed to 25 ppm of n-octane. The sensor’s response and recovery times were, respectively, 40 and 60 s. Drop coating of 41 Pd-doped SnO2 pastes was first introduced by . The sensing material was deposited on micromachined hotplates for the discrimination of CO, 6 NO2, and their binary mixtures. Briand et al. used this technique for the 42 deposition of 2% Pd-doped SnO2 paste on interdigitated Pt electrodes. The diameter of the drop was 400 mm with a thickness of a few tens of microns. It was deposited on a membrane of 1 1mm2 and 1 mm thick. The sensing material could be annealed on a chip using the sensor’s heater. For operating the device, a temperature of 300 C was reached with a power supply of 70 mW. The device showed a response of 2.2 and 1.4, respectively, to 10 ppm of CO and 2000 ppm of CH4. Despite their high thickness, drop coating has led to highly stable, reproducible sensors with very good sensitivity. These results led to the large-scale commercialization of drop-coated metal oxide gas sensors by AppliedSensor GmbH, Germany, for the automotive market.43 The microhotplate technology developed 6 by Briand et al. has been combined with much thinner optimized SnO2 Micromachined semiconductor gas sensors 429

(a)

(b) Sensing layer Heating electrodes

Sensing electrodes Figure 13.7 (a) SEM image of a drop-coated metal oxide gas sensor from AppliedSen- sor GmbH. (b) Three-dimensional schematic drawing of the sensor structure. From Blaschke M, Tille T, Robertson P, Mair S, Weimar U, Ulmer H. MEMS gas-sensor array for monitoring the perceived car-cabin air quality. IEEE Sens J 2006;6(5):1268e1308. and WO3 films, having a thickness of less than 5 mm(Fig. 13.7). Typical gas responses are displayed in Fig. 13.8. The metal oxide drop was then further reduced by using capillaries for its deposition and could reach a diameter of about 20 mm.44 Smaller hotplates can be thus used, leading to a potential further decrease in power consumption. Drop coating was then used by many other groups. Among others,45 and 46 later, from Morante’s group in Spain used it for the deposition of SnO2 and 47 BaSnO3.Espinosaetal. in Italy deposited drops made of 1% Pt-doped WO3,1%Pt-dopedSnO2,1%Pd-dopedSnO2, and 1% Au-doped SnO2 on a suspended microhotplate with a diameter of 80 mm. It required about 8 mW to reach an operating temperature of 400 C. As test gas, the sensing films were tested with ethylene, acetaldehyde, ethanol, and ammonia. A second technique widely used for the deposition of thick-film metal oxides on alumina substrates is screen printing. Looking at the success met by the thick drop-coated films, screen printing was reconsidered. Vincenzi 48 et al. screen-printed Pd-doped SnO2 paste onto micromachined microhotplates. The paste also contained a glass frit (a low melting temperature 430 D. Briand and J. Courbat

(a) 60 CO 50 NO2 40 Butanol 30 20 10 Concentration (ppm) Concentration 0 0 50 100 150 200 250 300 350 400 450 500 Time (min) (b) 10–2 Pd3 Pt02 10–3 U

10–4

10–5

Log conductivity (S) 10–6 0 50 100 150 200 250 300 350 400 450 500 Time (min)

Figure 13.8 Gas response (b) of 3% Pd-doped SnO2 (Pd3), 0.2% Pt-doped SnO2 (Pt02), and undoped SnO2 (U) to changes in concentrations of CO, NO2, and butanol (a). From Blaschke M, Tille T, Robertson P, Mair S, Weimar U, Ulmer H. MEMS gas-sensor array for monitoring the perceived car-cabin air quality. IEEE Sens J 2006;6(5):1268e1308. glass) to increase its viscosity and improve adhesion to the substrate. Partic- ular care had to be taken to avoid breaking the SiO2/Si3N4 membrane during film deposition. This was achieved by using a special stencil, which reduced pressure on the membrane. The sensing film was 250 350 mm2 and had a thickness of about 40 mm. The film was then fired at 650 C for 1 h, using the sensor’s heater. For gas detection, the devices operated at 400 C with a power of 30 mW and were evaluated with CO, CH4, and NO2. Fairly low responsesd1.2 for 50 ppm of CO, 1.03 for 1000 ppm of CH4, and 1.7 for 0.1 ppm of NO2dwere obtained. It was ascribed to the glass frit, which insulated the SnO2. To avoid breaking the membranes during screen printing,49 deposited a 5 mm thick, undoped SnO2 sensing film before releasing the membrane. It led to a significantly improved yield of 95% after encapsulation of the sensors. They showed responses of about 3e25 ppm ethanol and to 625 ppm of ammonia and 50 8e62.5 ppm of acetone. From the same group, screen printed SnO2 and WO3 pastes on micromachined transducers. When exposed to CO, Micromachined semiconductor gas sensors 431

low responses were obtained by SnO2 and no response was observed with WO3. In the case of an exposure to 1 ppm of NO2, responses of 3.63 with SnO2 at 250 C and 8.91 with WO3 at 200 C were measured. More- over, Ivanov et al. sputtered the same materials so as to investigate and compare the sensing properties of thin- and thick-film metal oxide layers. The results revealed that thick-film gas sensing layers have a higher degree of sensitivity than thin-film layers. This is due to the nature of the deposited film, which is more compact in the case of thin films, thus reducing the surface-to-volume ratio.18 SnO2 screen printing paste contains a binder to control the rheological properties and to ensure a good adhesion of the film to the substrate. Glasses bring problems of SnO2 percolation and thus reduce the conductivity. Remedy to this issue,51 evaluated different inks with an optional organic binder, instead of a mineral binder, and with Sn alkoxide, which lead to the formation of SnO2 during thermal annealing. Sensor films with a low conductance were obtained when no binder was used because of numerous cracks in the layer. The presence of both the organic binder and the alkoxide gave good results in terms of paste adhesion and conductivity, but the pattern resolution achieved was limited. However, nowadays, screen printing resolution down to 20 mm has been demonstrated in the field of printed electronics and better results could be expected for metal oxide pastes. Beside drop coating and screen printing, a further technologydflame spray pyrolysis (FSP)dshowed promising results. The deposition technique consists in spraying liquid precursors, which form a flame. The precursors react in the gas phase with the subsequent particle formation. This method allows a good control on morphologydamorphous or crystallinedas well as doping. Films with thicknesses of a few micrometers which do not require any annealing can be obtained. Sahm et al.52 used this method for the deposition of SnO2 on alumina substrates. Gas measurements were performed. The SnO2 sensing film showed a good response to low concentrations of NO2 (below 200 ppb) and propanal, and a low response to CO, which is 53 typical for undoped SnO2 films. Kuhne€ et al. used the same method for the deposition of Pt-doped SnO2 onto micromachined hotplates. The sensing film was patterned through a shadow mask. The transducer coated with the sensing film is presented in Fig. 13.9(a). The devices operated at 250 C with a power supply of about 25 mW. It showed a good response toward ethanol concentrations between 25 and 100 ppm, as illustrated in Fig. 13.9(b). 432 D. Briand and J. Courbat

(a) 100 μm

(b) 5

4 )

Ω 3

2 25 25 50 50

Sensor resistance (M 100

1 0 5101520 25 30 Time (min)

Figure 13.9 (a) SEM image of a Pt-doped SnO2 film deposited by flame spray pyrolysis; (b) gas response to ethanol. Values given are EtOH concentrations (ppm); 90% r.h. (20 C); Tm ¼ 250 C. (b). Adapted from Kuhne€ S, Graf M, Tricoli A, Mayer F, Pratsinis SE, Hierlemann A. Wafer-level flame-spray-pyrolysis deposition of gas-sensitive layers on microsensors. J Micromech Microeng 2008;18:035040.

13.4.3 Temperature modulation Metal oxide gas sensors can be operated in two modes: constant temperature (i.e., isothermal) and temperature-modulated modes. In constant temperature mode, the selectivity can be enhanced by using an array of sensors covered with different materials or dopants38,54; or by operating at different Micromachined semiconductor gas sensors 433 temperatures.16,41 However, the use of several sensors considerably increases the complexity and the power consumption of the system. Additionally, a drawback with constant temperature operation is that a mixture of oxidizing and reducing gases can offset each other and no signal variation will be observed.18 With the micromachining of the devices, their thermal response times were drastically reduced to the millisecond range. This allowed their operation in a pulsed or cycled temperature mode to avoid the interference of humidity and allowed the discrimination of several gases with one single sensor. This measurement technique was first introduced by55. They applied a sine signal to the sensor heater and measured the response of the SnO2 gas sensing layer when exposed to different analytes. They observed that methane and propane gave a higher response with a heater at its maximum temperature, while CO is better measured in a cooling state. Each gas can be identified by a specific temporal response pattern, which depends on its chemical reaction with the gas-sensitive material.56 Major investigations related to temperature-modulated micromachined metal oxide gas sensors were performed in Semancik’s group. Ratton et al.57 applied a sawtooth signal shape to the heater to reach temperatures up to 550 C. The behavior of methanol, ethanol, acetone, and formaldehyde was studied. The sensor signal was processed through the Grame Schmidt approach, fast Fourier transform (FFT), Haar wavelet transform, or the Granger approach to reduce the number of coefficients describing the signal and to retain as much relevant information as possible. Best results were achieved with the Haar transform, which efficiently compressed the information while removing noise and drift effects. Kunt et al.58 used the same device to discriminate methanol and ethanol using temperature modulation. Both gases responded differently to the temperature change, as can be seen in Fig. 13.10. In this study, they optimized the temperature profile to improve response selectivity between these two gases. The sensitivity can be further improved by taking advantage of the unsteady state of the number of oxygen species at the surface of the metal oxide when its temperature is changing. Llobet et al.59 showed that the transient response of thermally cycled metal oxide sensors decreases the sensor’s response to humidity and to the drift in the resistance of the gas- sensitive layer. Several options of temperature variations have been presented in the literature to improve selectivity. Different waveforms at different frequencies have been applied to the heater of the gas sensor to achieve thermal cycling of its temperature. The sensor response can be then analyzed by signal processing. FFT was used by60. They applied a 434 D. Briand and J. Courbat

(a) 2

–1

–2

Normalized conductance 96 98 100 102 104 106 108 110 Time (s) (b) 380 360 340 320 300 280 Temperature (°C) Temperature 260 96 98 100 102 104 106 108 110 Time (s)

Figure 13.10 (a) Conductance response of a SnO2 gas sensing layer to ethanol (dashed line) and methanol (solid line) for temperature pulses of 275 and 380 C (b). From Kunt TA, McAvoy TJ, Cavicchi RE, Semancik S. Optimization of temperature programmed sensing for gas identification using micro-hotplate sensors. Sens Actuators B 1998;53:24e43 sine wave and its second harmonic to the sensor heater to improve the selectivity of a SnO2 semiconductor gas sensor. Depending on the phase shift of the second signal compared with the first, discrimination between alcohols, hydrocarbons, and aromatic compounds could be performed. Fig. 13.11 shows the sensor response to ethanol, ethane, and toluene as representative examples of these gas families. Llobet et al.59 used discrete wavelet transform and an artificial neural network to measure and discriminate CO, NO2, and their mixture. The wavelet technique gave better results than FFT in terms of data compression and tolerance to noise and drift in the sensor response. A system based on simpler electronics relies on pulsing the temperature (i.e., the heater is only switched on and off). Depending on the duty cycle, it allows a significant reduction in power consumption.44 Among other techniques, this was used by Faglia et al.11 for the detection of CO with an Au-doped SnO2 film. They used a square signal with a period between 0.5 and 180 s. The heater was powered for 100 ms, which was sufficient to reach a steady state. Beside a reduction in power consumption, Faglia et al. Micromachined semiconductor gas sensors 435

(a) 1.0 1.0 1.0 1.0 (a-1) (a-2) (a-3) (a-4)

0.5 0.5 0.5 0.5

400 450 500 550 400 450 500 550 400 450 500 550 400 450 500 550 (b) 0.4 1.4 1.4 1.4 (b-1) (b-2) (b-3) (b-4)

0.2 0.2 0.2 0.2 (mS) G

400 450 500 550 400 450 500 550 400 450 500 550 400 450 500 550 (c) 1.0 1.0 1.0 1.0 (c-1) (c-2) (c-3) (c-4)

0.5 0.5 0.5 0.5

0 0 0 400 450 500 550 400 450 500 550 400 450 500 550 400 450 500 550 T (K) Figure 13.11 Sensor temperature versus sensor conductance of the sensor when a sine wave was applied to the heater without the second harmonic (dashed line) and with the second harmonic (solid line) with a phase shift of (1) 0 rad, (2) p/2 rad, (3) p rad, and (4) 3p/2 rad. The sample gases tested were 1000 ppm of (a) ethanol, (b) ethane, (c) toluene. From Nakata S, Okunishi H, Nakashima Y. Distinction of gases with a semiconductor sensor under a cyclic temperature modulation with second-harmonic heating. Sens Actuators B 2006;119:556e561. observed an increase in sensor response, compared with DC measurements, for periods up to 20 s. Therefore, such a method can allow a reduction in power consumption while improving sensing performances. 13.4.4 Packaging Silicon micromachined semiconductor gas sensors are mainly packaged using standard metallic transistor outline (TO) headers as support, and wire bonding is used for their electrical connection. Typically, a metallic cap with a grid is ﬁxed to the TO header with a hydrophobic gas permeable membrane on top of it. A ﬁltering agent can be also included in the package. The use of silicon microfabrication techniques brings not only the ability to process the sensors at wafer level but also, as demonstrated in Raible et al.61 in 2006, the encapsulation and testing of the sensors at wafer level. This concept allows liquid-tight sealing of gas sensor devices, which protects 436 D. Briand and J. Courbat

(a) Thick-film SnO2 layer Diffusion filter membrane Pyrex filter support Optional base to ease pick and place Micro-machined substrate Micro-machined hotplate membrane Bonding pads

(b)

Figure 13.12 (a) Diagram of the wafer-level packaged metal oxide sensor; (b) optical picture of an individual sensor area with the Pyrex rim and the metal oxide drop before the fixation of the gas permeable membrane. From Raible S, Briand D, Kappler J, de Rooij NF. Wafer level packaging of micromachined gas sensors. IEEE Sens J 2006;6(5): 1232e1235. them during production (e.g., wafer dicing) and later in the application, while still allowing the target gases to reach the sensing layer. The basis of wafer-level packaging is the combination of a structured Pyrex wafer with a micromachined substrate wafer. Thereafter, thick-film SnO2 layers are deposited and stabilized before a diffusion membrane is attached, which seals the wafer stack as shown in Fig. 13.12. The wafer stack is finally diced into individual sensor elements which can be mounted on printed circuit board using different interconnection methods, such as chip on board, flip-chip, tape-automated bonding, and so on (Fig. 13.13). Briand et al.62 reported on a higher level integration of wafer-level packaged micromachined metal oxide gas sensors. The concept was based on the insertion of the metal oxide drop into the micromachined cavity in the silicon substrate with the platinum electrodes at its bottom. Using this Micromachined semiconductor gas sensors 437

Figure 13.13 Chip on board wafer-level packaged metal oxide gas sensors on printed circuit board. From Raible S, Briand D, Kappler J, de Rooij NF. Wafer level packaging of micromachined gas sensors. IEEE Sens J 2006;6(5):1232e1235. approach, the Pyrex rim was no longer necessary and the gas permeable membrane could be fixed directly onto the silicon substrate to close the cavities containing the drop-coated metal oxide film (Fig. 13.14). For a 200 mm-wide deep reactiveeion-etched (DRIE) membrane, a power consumption of 15 mW was reached at 300 C. DRIE technology also allows the reduction of the chip size to a minimum, compared with KOH etching. Following the trends in the field of sensor packaging and mounting, surface-mount devices are appearing on the market using a plastic, molded package as a cost-effective approach, as it is described for the gas sensor product from Sensirion AG, Switzerland, in Section 6.9.

13.5 Complementary metal oxide semiconductorecompatible metal oxide gas sensors CMOS-compatible and SOI-based microhotplates used as transducers for metal oxide gas sensors were reported, respectively, by Suehle et al.63 and Laconte et al.64 They addressed the realization of the hotplates themselves in a CMOS-compatible process with an integrated poly-Si heater. But the real benefit of this technology comes with the integration of the complete driving and readout electronics on the sensor chip. Beside the potential reduction of power consumption and the cost of the sensor system, the number of bonding wires can be decreased, as can the packaging. The integration of the electronic circuitry can also improve signal response fidelity due to on-chip signal processing and amplification and conditioning of small 438 D. Briand and J. Courbat

(a) Gas permeable membrane

200–600 μm 390 μm Si Si

SnO2 SiO2

Si3N4 Electrodes Heater

(b)

(c)

Figure 13.14 (a) Diagram of the wafer-level packaged semiconductor gas sensors with the metal oxide film deposited in the silicon micromachined cavity; (b) and (c) optical images of the DRIE etched 200 mm-wide drop-coated metal oxide device (b) before and (c) after encapsulation at the wafer level. From Briand D, Guillot L, Raible S, Kappler J, de Rooij NF. Highly integrated wafer level packaged MOX gas sensors. Proceedings of the Transducers’07 conference. Lyon, France, June 10e14; 2007. p. 2401e2404. Micromachined semiconductor gas sensors 439 sensor signals. Benefits can be brought to the operation of the sensor by allowing the implementation of driving, signal conditioning, and compensation strategies. However, if the yield of the formation of the sensing layer on the sensor chip is not sufficiently high, the failure cost will be significantly higher, together with the loss of the electronics. Four main concerns need to be addressed when integrating metal oxide sensors in a CMOS-compatible process: • The dielectric membrane of the microhotplates will be composed of CMOS dielectric films. It can be formed through a silicon micromachining postprocess either on the back or the front. • The standard electrically conductive materials are doped polysilicon and aluminum, which are not suitable to be used as heaters (Section 6.3.3) or electrodes (oxidation of Al) for the sensor. Implementing platinum, the commonly used material, as heater and electrode material involves postprocessing steps. Another approach for the heater is to use tungsten which can be available in CMOS technology. • The postdeposition of the metal oxide sensing layer needs to be CMOS compatible, and its postdeposition annealing is limited in terms of temperature and time. • Once the CMOS metal oxide sensor chip is available, the miniaturization of the device brings different issues to the CMOS electronics design. We refer the reader to the comprehensive review published by Gardner et al.65; for more information about electronics circuitry design. Afridi et al.66 have reported on an array of four bridge-type front micromachined hotplates with postprocessed gold electrodes and including interface electronics. The metal oxide films, tin oxide and titanium dioxide, were deposited using an LPCVD process when operating the microhotplates at different temperatures. A decoder was used to select a given microheater and sensing resistive layer, with a bipolar transistor or a MOSFET switch, respectively. The signal-to-noise ratio was improved using an on-chip operational amplifier. ETH Zurich, in Switzerland, has extensively developed CMOS- compatible metal oxide gas sensors with on-chip integrated circuitry.12,67 Postprocessing was used to include platinum electrodes on the hotplate coated with a drop-coated Pd-doped tin oxide film. Annealing of the metal oxide film was performed at a maximum temperature of 400 C, which prevented any degradation of the device. Fig. 13.15 presents an advanced analog/digital monolithic sensor system.17 Its fabrication was performed using an industrial CMOS process followed by postprocessing steps for 440 D. Briand and J. Courbat

(a) Reference resistor

Temperature sensor

n-well Si-island

100 μm (b) Ring Electrodes heater Nanocrystalline SnO2 Dielectric thick-film layer membrane

Temperature sensor Si-island Bulk-chip temperature sensor Analog Digital circuitry circuitry and (controller (c) converters and interface)

500 μm

Micro-hotplate

6.8 4.7 mm2 Figure 13.15 Monolithic metal oxideebased gas sensor system in CMOS technology: (a) close-up of the microhotplate (left) and SEM micrograph of a microhotplate with a drop-coated nanocrystalline thick-film layer (right); (b) cross-sectional diagram of the heated area of the sensor chip; (c) micrograph of the CMOS-based overall sensor system chip featuring microhotplates and circuitry.17 Micromachined semiconductor gas sensors 441 the patterning of the platinum electrodes, the release of the membrane by silicon micromachining, and the deposition of the sensing layer. Dielectric thin films available in the CMOS process were used for the thermally insulated membrane, electrical insulation, and passivation. The active area featured a circular-shape resistive heater, a temperature sensor, and electrodes to contact the sensing layer. In Fig. 13.15(c), the microhotplate, the analog circuitry (including analog-to-digital and digital-to-analog converters), and the digital circuitry are distinguishable. The digital part included a programmable digital temperature controller and a digital interface. This enabled control of the sensor temperature, as well as a readout of the temperature of the hotplate and the gas sensor signal. A logarithmic converter connected to the resistance layout of the sensitive layer not only allowed a first-order signal linearization but also helped to address the large variation range of the metal oxide resistance from 1 kU to 100 MU. A stand-alone version of the monolithic sensor system (including three transistor-heated microhotplates32 with fully digital temperature controllers and a digital interface) was developed to take complete advantage of this technology. In 2017, Sensirion AG, Switzerland, has released a CMOS compatible metal oxide gas sensor product for which more details can be found in Section 6.9. Robust high-temperature tungsten-based SOI microhotplates were reported by Ali et al.8 and have been successfully commercialized by Cambridge CMOS Sensors Ltd. in the United Kingdom. The hotplates are fabricated using a standard SOI CMOS process in a commercial foundry, followed by a DRIE postprocessing step to release the dielectric silicon dioxide closed-type membrane. The process was performed on 150 mm SOI wafers with a 0.25 mm-thick silicon device layer sitting on a 1 mm-thick box oxide layer used as etch stop during the DRIE of silicon. The silicon device layer is very thin and can be removed from the whole membrane area for better thermal insulation. One of the tungsten metal layers was used as heater and exhibited very stable behavior at a high temperature of 500 C. An ultralow power consumption of 12 mW and a fast transient time of 2 ms to reach 600 C were reported. Fig. 13.16 presents a diagram of this device. The complete integration of the CMOS electronic circuitry with the sensor element is still to be demonstrated. 442 D. Briand and J. Courbat

Passivation Gas Silicon heat Silicon Tungsten Metal heat sensing spreading dioxide plate spreading material plate heater PMOS NMOS

Buried silicon P N PPN N dioxide

Substrate

Sensor CMOS Figure 13.16 Design of tungsten SOI chip: gas sensor and integrated CMOS circuitry. From Ali SZ, Udrea F, Milne WI, Gardner JW. Tungsten based SOI microhotplates for smart gas sensors. J Microelectromech S 2008;17(6):1408e1417.

13.6 Micromachined field-effect gas sensors The field-effect gas sensing principle was first demonstrated by Prof. Lundstrom€ in 1975 by replacing the standard aluminum gate of a MOSFET with a catalytic metal, such as palladium, for the detection of hydrogen.68 By heating up the device, hydrogen molecules dissociate in hydrogen atoms, which diffuse through the catalytic metal, reaching the metaledielectric interface of the FET devices. Electric dipoles are created, which induce a change in the IeV curve characteristics of the FET device. By tuning the catalytic gate material of the device, a series of gases (mainly containing hydrogen atoms) can be sensed using the FET as a transducer.69 Extensive literature can be found on the topic and AppliedSensor GmbH is now commercializing the technology mainly for application in the fuel cell market. Modulating the temperature is also of interest for this sensing principle, and some work has been undertaken in that direction. However, low power and low thermal mass devices are desirable for this purpose.70 These devices have also been developed on silicon carbide for applications in harsh environment.71 At the end of the 1990s, in the framework of the European project Chemical Imaging for Automotive Applications (CIA), reducing the power consumption of GasFETs was identified as being of interest to the automotive market. Developments have been undertaken by Briand et al.72 to achieve the thermal insulation of a GasFETs array based on the microhotplate concept. At that time, the technology was further developed for its integration into an electronic nose by Nordic Sensors Technologies, Sweden (now AppliedSensor). Micromachined semiconductor gas sensors 443

(a) Dielectric membrane (0.5 μm thick) LPCVD + PECVD nitride 1.8 mm Al

MOSD MOS Si 400 m Si μ 900 μm 10 μm

MOS MOSFET DDIOD E

(b) Heater

MOSFET

Ground

MOSFET

Diode

MOSFET

Heater Figure 13.17 (a) Diagram of cross-sectional view of the low-power MOSFET array gas sensor. The electronic components are located in a silicon island insulated from the silicon chip frame by a dielectric membrane made of two nitride layers; (b) Optical image of the low-power micromachined MOSFET gas sensor showing the electronic devices in the silicon island suspended by a dielectric membrane. From Briand D, van der Schoot B, de Rooij NF, Sundgren H, Lundstrom€ I. A low-power micromachined MOSFET gas sensor. J Microelectromech S 2000b;9(3):303e308

Basically, using silicon micromachining, an array of four GasFETs devices, with different catalytic layers (Pd, Ir, Pt), were located on a silicon island thermally insulated from the silicon chip frame by a thin-ﬁlm dielectric membrane made of silicon nitride;73 Fig. 13.17. A two-step wet silicon anisotropic etching in KOH was developed to achieve a 10 mm-thick silicon plug underneath the dielectric membrane, in which the electrical components were located. A doped silicon resistor used as heater and a diode used as a temperature sensor were integrated into the design, as shown in Fig. 13.17. 444 D. Briand and J. Courbat

Processing, however, remained heavy, with many photolithographic steps. Power consumption was significantly reduced to 90 mW for an operational temperature of 170 C. But the most interesting feature was the fast modulation of the temperature. A thermal time constant of less than 100 ms could be reached with sensing devices configured in this way. Modifications of the kinetics of the gas reactions with the sensing film occurred when modulating the temperature. They depended on the sensor “history,” on the nature of the gaseous atmosphere, and on the type of materials used as the catalytic film. Reduction of the recovery time of the device was achieved by performing a temperature pulse following the gas exposure, and the discrimination of gases in a mixture using temperature cycling (100e200 C) was especially valuable, with an effective resolution at a temperature modulation of “low” frequency (0.1 Hz) and large amplitude.24,25 The data were Fourier transformed before the evaluation was made using principal components analysis plots. Discrimination was shown for gaseous mixtures of hydrogen and ammonia (10e100 ppm) in air (Fig. 13.18). (a) (b) 2.04 1.965

2.03 1.96

2.02 1.955

2.01 1.95 Sensor signal (V) Sensor signal (V) 2 1.945 0 50 100 150 200 0 50 100 150 200 Time (s) Time (s) (c) (d) 1.75 1.73 1.74 1.72

1.73 1.71

1.72 1.7 Sensor signal (V) Sensor signal (V) 0 50 100 150 200 0 50 100 150 200 Time (s) Time (s) Figure 13.18 Ir-MOSFET voltage (continuous line) at a temperature variation between 150 and 200 C at 0.01 Hz (dashed line) in the presence of (a) pure synthetic air; (b)

100 ppm H2; (c) 10 ppm NH3; (d) 10 ppm NH3; and 100 ppm H2. From Briand D, Tom- assone G-M, de Rooij NF. Accelerated ageing of micro-hotplates for gas sensing applications. Proceedings of IEEE sensors 2003 conference. (Toronto, Canada); 2003a. p. 1314e1317; Briand D, Wingbrant H, Sundgren H, van der Schoot B, Ekedahl L-G, Lundstrom€ I, de Rooij, N. F. Modulated operating temperature for MOSFET gas sensors: hydrogen recovery time reduction and gas discrimination. Sens Actuators B 2003b;93:276e285. Micromachined semiconductor gas sensors 445

13.7 Nanostructured gas sensing layers on microhotplates Nanowires are seen as a solution with which to improve the sensitivity, selectivity, stability, and response time of metal oxide gas sensors. Meier 74 et al. grew SnO2 nanowires of 100 nm in diameter by the vaporesolid growth method. For testing, they were deposited onto micromachined hotplates and contacted with a focused ion beam scanning electron microscope (FIB-SEM), as shown in Fig. 13.19. Because of their diameter being similar to the Debye length, a completely depleted conduction channel can be obtained. Maximum response to CO and NH3 occurred at about 260 C. 75 SnO2 nanoparticles can also be grown by solegel method. Li et al. achieved SnO2 nanomaterial by precipitating SnCl4$5H2O from an aqueous solution.

(a)

100 μm

Four electrodes heater Polysilicon Etched pit NW

(b)

SnO2 nanowire

Pt FIB contacts

500 nm Figure 13.19 (a) Optical image of microhotplates with nanowires, NW. (b) SEM images of SnO2 nanowire on a microhotplate. From Meier DC, Semancik S, Button B, Strelcov E, Kolmakov A. Coupling nanowire chemiresistors with MEMS microhotplate gas sensing platforms. Appl Phys Lett 2007;91:063118. 446 D. Briand and J. Courbat

The obtained powder could then be doped by adding TiO2 or carbon nanotubes. The nanopowders were deposited on microhotplate membrane. A droplet of deionized water was first drop coated on the membrane. It was followed by scattering SnO2-based powder on the substrate. The powder was mixed with water to obtain a paste, which was later dried. Sensing films of approximately 200 mm in thickness were produced. The sensors were tested against ethanol at 300 C. They showed, however, a poor selectivity 76 toward methanol, acetone, formaldehyde, NH3, and toluene. Similarly, obtained Au-doped SnO2 nanocomposites. They first precipitated SnCl2$2H2O to get SnO powder. It was then mixed to HAuCl4 to obtain Au nanoparticles attached to the surface of SnO2 mixture of nanoparticles and nanowires. The latter was maskless deposited by DPN (dip-pen nanolithography), which allowed confining the sensing material to the electrode area of a commercial microhotplate. Concentrations of ethanol between 100 and 1000 ppm could be detected at 400 C. The sensor revealed, however, to be sensitive to humidity and showed fair selectivity toward toluene and acetone. Materials other than SnO2 also exhibited good gas sensing performances. 77 Ryu et al. fabricated In2O3 nanowires by a laser ablation method. The nanowires were then sonicated in isopropanol to obtain a suspension, which was deposited onto microhotplates. When operating at 275 C, responses (R/R0) of 1.6e50 ppm of ethanol, of 2e100 ppm of CO, and of 0.5e50 ppm of H2 were measured. In addition, the micromachined gas sensor exhibited a short gas response time of about 22 s. Vapor phase growth is a technique that can be used for producing rather high quantity of nanomaterials. Marasso et al.78 used it to form ZnO nano- tetrapods from a metallic Zn seed. The ZnO nanostructures were dispersed in a solvent before their precipitation on the membrane of a hotplate by centrifugation. The deposited structure exhibited a good adhesion to the substrate avoiding any firing process. The obtained sensors revealed a maximum response to ethanol and methane at 400 C and to H2S and NO2 at 300 C. An alternative method for growing nanotubes is through hydrothermal process. It involves crystallizing material from an aqueous solution at temperature typically between 80 and 90 C. Such method was used by Shao et al.79 to obtain ZnO nanowires from a ZnO seed layer. Their diameters were between 50 and 300 nm for a length of about 6 mm. They were then drop coated onto a commercial microhotplate. An AC signal was applied between electrodes to align the nanowires. A subsequent annealing was performed at 400 C. They showed good response to NH3 when Micromachined semiconductor gas sensors 447 heated at 350 C. Lee et al.80 obtained ZnO nanowires on a microhotplate through a lift-off process. A photoresist mask was patterned and the substrate was immersed in an aqueous solution for hydrothermally growing the nanowires. Once the process was completed, the photoresist was stripped. Chen et al.81 grew ZnO nanowires in situ on the electrodes of a microhotplate. Zinc acetate was first drop coated onto the electrodes. After drying, a seed film of zinc acetate crystallites was formed. It was followed by hydrothermal process to grow grass-like nanowires. They could be then used as seed layer for a second hydrothermal process to obtain branch structures onto them. These nanostructures showed a very good sensitivity toward H2S when heated at 300 C with a limit of detection of 3 ppb. Other materials can be grown by hydrothermal processes. For instance,82 obtained hexagonal WO3 nanorods of 80e150 nm in diameter and 4e5 mm in length from sodium tungstate. The obtained nanowires could be decorated with Au or Pt nanoparticles. Au-doped nanowires had an enhanced sensitivity toward H2S with a concentration detection as low as 5 ppb. Doping additionally reduced response time to 1 ppm of H2S compared with undoped WO3 wires from 300 s down to 30e40 s. Inkjet printing can be used to pattern hydrothermally grown nanowires, thus avoiding shadow-masking or photoresist patterning. Krainer et al.83 deposited a suspension of WO3 nanowires with a commercially available inkjet printer on microhotplate membrane. Once deposited, the deposited droplets were annealed at 400 C for 12 h. Sub-ppm concentrations of H2S could be detected at 250 C independently of the relative humidity level. Nanotubes can also be grown by CVD processes. Recently84, showed that AACVD (Aerosol-Assisted Chemical Vapor Deposition) technique was suitable for growing WO3 nanoneedles. The nanoneedles could be functionalized with Au and/or Pt nanoparticles. The method involves temperatures between 350 and 600 C, which are compatible with MEMS-based devices. The patterning is typically made through a shadow mask. The fabricated sensors showed good discrimination between ethanol, hydrogen, and CO when heating between 100 and 300 C. These gases are of particular relevance in proton-exchange fuel cells. AACVD was also reported to be used for growing Cu2O-decorated WO3 nanoneedles by Annanouch et al.85 in one-step process on microhotplate. The resulting sensor showed a response of 27.5 to 5 ppm of H2S when heated at 390 C with a limit of detection of approximately 300 ppb. In addition, the device exhibited a selectivity against H2, CO, NH3,C6H6, and NO2. The same author reported later PdO nanoparticle-decorated WO3 448 D. Briand and J. Courbat

Figure 13.20 WO3 film morphology on a micromachined hotplate observed by SEM images at low (a and b) and high (c) magnification. (d) Cross section of WO3 nanoneedles. Reprinted with permission from Annanouch FE, Haddi Z, Ling M, Di Maggio F, Vallejos S, Vilic T, Zhu Y, Shujah T, Umek P, Bittencourt C, Blackman C, Llobet E. Aerosol-asssited CVD-grown PdO nanoparticle-decorated tungsten oxide nanoneedles extremely sensitive and selective to hydrogen. ACS Appl Mater Interfaces 2016;8:10413e10421. Copyright (2016) American Chemical Society. nanoneedle with a two-step AACVD process.86 Their integration on a microhotplate is illustrated in Fig. 13.20. It aimed at H2 detection in renew- able energy source. Exposure to 500 ppm of H2 led to a sensor response of 1670 when heated at 150 C. The sensor response was defined as the ratio of the sensor resistance in air to the analyte of interest for reducing gases and the opposite for oxidizing gases. The response decreased above that temperature and provided unreproducible results. Additionally, the sensor had a good selectivity against NH3,C6H6, and CO. Nanowires can be grown directly from a substrate. For instance,87 grew CuO nanowires directly from 600 nm-thick Cu structures placed on the electrodes of a microhotplate. The latter was heated at approximately 335 C using its buried heater. Growth occurred in a gas test chamber with synthetic air. This process resulted in 1 mm long nanowires with a diameter of about 20 nm. As the sensors were mounted on PCB, nanowire growth could be electrically monitored as well as CO sensing capabilities. Because gas measurement occurred in the very same chamber, the sensors Micromachined semiconductor gas sensors 449 could be assessed without being exposed to ambient environment. They showed responses (R/R0*100) of 6.4% and 27.6% to CO concentrations of, respectively, 1 and 30 ppm when operating at 325 C. The sensor performances dropped once exposed to humid environment because of hydroxylation of the CuO surfaces. A main issue toward reducing the power consumption of metal oxide gas sensors is their operating temperature, which is reduced in some cases by using nanostructures. Previous examples used microhotplates to reach the optimum thermal operating conditions. In an alternative move,88 addressed this problem by directly using the probing current applied to the nanowires as the heat source. This significantly simplified the device by avoiding the need for the integration of a heater into the hotplate. Moreover, it reduced the heated area and, consequently, power consumption. Currents in the range of 0.1e300 nA were flowing through an SnO2 nanowire to heat it up to 300 C. The measured power consumption was 30 mW, two to three orders of magnitude lower than “standard” micromachined metal oxide gas sensors, making them compatible with energy harvesting systems. Very fast sensors were obtained with response times in the millisecond range. They had a good response to CO and NO2. Si nanowires were used as gas sensors by89. The devices could operate at room temperature, drastically reducing their power consumption. They could be thus transferred on polyethylene terephthalate (PET) plastic foil as substrate (Fig. 13.21). A response of about

50 μm Figure 13.21 SEM image of an array of SNAP nanowire sensors. Each device (horizontal strip) is contacted by two Ti electrodes (oriented vertically) that extend to larger pads (top and bottom image edges). Inset: digital photograph of the ﬂexible sensor chip. From McAlpine MC, Ahmad H, Wang D, Heath JR. Highly ordered nanowire arrays on plastic substrates for ultrasensitive ﬂexible chemical sensors. Nat Mater 2007;6:379e384. 450 D. Briand and J. Courbat

2 was obtained under an exposure of 2 ppm of NO2. The detection of NO2 concentrations as low as 20 ppb was possible. The device response time was up to few minutes, depending on the gas concentration. Purge cycles with vacuum and fresh air were necessary for the sensor to recover after an exposure to NO2. The nanowires could be functionalized with alkane-, alde- hyde-, and amino-silane to improve selectivity and allow differentiation of a binary mixture of acetone and hexane. The fabrication of nanowires has been mastered and they have shown to be suitable for gas sensing. However, several issues remain for their large-scale use in commercial devices and for the achievement of reproducible results. It mainly concerns the precise location of the nanowires on a speciﬁcareaand their electrical contact. From an operation point of view, to beneﬁtfromtheir low operational temperature for gas detection, sensitivity to humidity and slower desorption kinetics will need to be addressed in some cases.

13.8 Semiconductor gas sensors on polymeric foil and their additive manufacturing 13.8.1 Semiconductor gas sensors on polymeric foil The use of plastic substrates, since 2008, has been seen as a solution to further decreasing sensor cost and manufacturing complexity, compared with devices manufactured on silicon or ceramic substrates. Plastic additionally shows other benefits, such as compatibility with large-scale fabrication (roll-to-roll), printing compatibility, lightweight, and conformality. Such devices aim at new applications where low cost is a prerequisite: smart sensing labels, wearable devices, consumer goods, distributed systems, and so on. However, metallic oxide films are usually annealed at high temperature, and the main challenge of processing them and operating them on plastic substrates is the limited thermal budget. Nanowires, the FSP deposition technique, and low sintering temperature nanoparticle inks are potential candidates for integration at a relatively low temperature onto polymeric transducing platforms of performing metal oxide materials. Briand et al.90 were the first to demonstrate the use of polyimide (PI) as a substrate for the fabrication of plastic-based metal oxide gas sensors. Two types of devices were fabricated by standard microfabrication equipment. The first solution consisted in using silicon as the substrate, which was spin coated with a PI layer. Once the bulk silicon was dry etched, a PI membrane embedding a Pt-based heater and with electrodes on top was released. The second solution was based on the use of a commercially available PI foil Micromachined semiconductor gas sensors 451 as the substrate. A Pt heater was patterned and covered with a photosensitive spin-coatable PI layer used as a dielectric film to electrically insulate the electrodes on top. In both configurations, the interdigitated electrodes were drop coated with a Pd-doped SnO2 layer as the gas sensing film with a maximum annealing temperature at 450 C. These devices showed good gas sensing performances but suffered from excessive power consumption when operating at 325 C: 82 mW for devices on silicon and 130 mW for the device on PI foil. To reduce power consumption,9 investigated the miniaturization of drop-coated metal oxide gas sensors on PI foil. Their transducers were optimized in terms of power consumption and temperature uniformity through electrothermal simulations. Devices from 100 mm down to 15 mm were produced. With the idea of reducing power consumption further, the PI foil could be dry etched in an O2/CF4 plasma to obtain closed and suspended membranes about 3 mm thick. The deposition of the 44 metal oxide layer (Pd-doped SnO2) was carried out with micropipettes. The smallest droplet had a diameter of 20 mm(Fig. 13.22(a)). A power consumption as low as 6 mW was required to reach 300 C with a 15 mm-wide heater with a closed membrane in a continuous operating mode. With a simplified fabrication process avoiding the bulk micromachining of the PI foil, only 10 mW was necessary with a heater of the same size. These sensors could operate for more than 1 year at 200 C.91 The sensors worked in both continuous and pulsed modes, which decrease the power consumption to the sub-mW level. The devices showed to be effective for the detection of CO (Fig. 13.22(b)), CH4, and NO2. Furthermore, a method for the encapsulation of chemical sensors at foil level was demonstrated.92 It consisted in a prepatterned rim made of a dry photoresist film laminated onto the PI substrate containing the gas sensors. They were covered with a water-repellent gas permeable membrane. ZnO nanowires were grown on PI-based microhotplates by93. Zn was sputtered onto the substrate through a shadow mask and then oxidized for 12 h at 300 C. Such a relatively low temperature was required to avoid damaging the plastic foil. The ZnO nanotubes showed a response toward 89 NO2. PET foils were used by McAlpine et al. as the substrate onto which nanotubes were deposited (see Section 6.7 for more information). The device showed itself to be suitable for measuring NO at room temperature. The operation of metal oxide gas sensors on plastic foil was successfully demonstrated. However, to make them fully compatible with large-scale fabrication techniques, i.e., printing, additional work is required. This topic is addressed in the next section. 452 D. Briand and J. Courbat

(a)

Electrode 20 μm connections

Gas sensing layer

Heater 50 μm connections

(b) 6 100 μm - 28 mW 5 50 μm - 18 mW 25 μm - 11 mW 4 15 μm - 10 mW

2 Chemresistive response (—) Chemresistive

1 20 30 40 50 60 70 80 90 100 CO concentration (ppm) Figure 13.22 (a) Optical image (top view) of metal oxide gas sensor on PI; (b) gas response to CO for several sensor sizes when operating at 250 C. Adapted from Courbat J, Briand D, Yue L, Raible S, de Rooij NF. Drop-coated metal-oxide gas sensor on polyimide foil with reduced power consumption for wireless applications. Sens Actuators B 2012;161:862e868.

13.8.2 Printing semiconductor gas sensors Recently, since 2010, with the emergence of printing techniques, new deposition methods compatible with large area manufacturing have been applied to gas sensing materials. Inkjet-printed pure and doped SnO2 was performed on silicon and alumina substrates.94 The use of inkjet printing facilitated doping by the consecutive printing of SnO2 and a dopant. A pure SnO2-based sensor exhibited a response of about 7e50 ppm of ethanol and 55 when exposed to 50 ppm of H2S when operating, respectively, at 425 and 179 C. However, their printed layers required annealing at Micromachined semiconductor gas sensors 453

550 C, making them incompatible with plastic substrates. This drawback was counteracted by Peter et al.95, who developed a titanium-doped chromium oxide ink that did not require any firing. The adhesion to the silicon substrate and the film stability was improved by sintering the printed layer at 400 C. This temperature is, however, compatible with a high performance polymer such as some PIs. Moreover, being an additive technique, inkjet printing is of significant interest with regard to the local patterning of different sensing films on one substrate. In the case of arrays, all sensing material can be deposited simultaneously, simplifying fabrication of the device. Kukkola et al.96 used another technique compatible with roll-to-roll processing: gravure printing. They deposited WO3 sensing films on interdigitated electrodes patterned on Kapton HN PI foil from DuPont. However, the fabrication of an integrated heating element was not addressed in this study. For gas response measurement, the sensor was placed in a heated gas cell at 200 C. A gas response was obtained for a concentration of 5 ppm of NO. A coplanar architecture was reported by Ramírez et al.97 in 2018 to implement in one single layer the electrodes and the heating element of printed gas sensors. The design includes two electrodes and three contacts. One of the electrodes works as heating element and, simultaneously, drains the sensing current. Compared with other coplanar topologies, this approach simplifies the transducers processing to a single printing step, avoiding the use of an interdielectric layer between heater and electrodes. This cost-effective architecture and process was applied to the fabrication of heated transducers for metal oxide gas sensors. The two electrodes were made by inkjet printing of gold on PI foil. For the validation of the concept, a Pt-loaded WO3 sensing layer was grown on top of these transducers printed with the proposed topology. This simple architecture has strong potential for the realization of fully printed resistive gas sensors and can be implemented as well in cleanroom processed transducers. The first fully inkjet-printed tin dioxide (SnO2) gas sensor was reported by Rieu et al.98 in 2016. Gold electrodes and heater were inkjet printed on each side of a PI substrate. A SnO2-based solegel ink was inkjetted onto the electrodes. A final annealing at 400 C allowed to synthetize the SnO2 sensing film. The device was operated at temperatures between 200 and 300 C using the integrated heater. The proper operation of the fully printed metal oxide gas sensors was validated under exposure to carbon monoxide and nitrogen dioxide, in dry and wet air. In 2018, Khan et al. have reported on a low-power metal oxide gas sensor using aerosol jet printing to reduce 454 D. Briand and J. Courbat the area of the hotplate transducer to 500 500 mm2. Aerosol jet was used to print the gold heater and electrodes and the interdielectric layer made of PI. The transducer consumes 78 mW at an operating temperature of 200 C. Inkjet printing was used to coat the transducers electrodes with Pd-doped tin dioxide nanoparticles.

13.9 Manufacturing, products, and applications Large volume manufacturing of semiconductor gas sensors has begun in the early 2000s with the company MiCS (MicroChemical Systems SA) in Switzerland, now part of the SGX Sensortech group, and AppliedSensor GmbH in Germany, bought by AMS AG in Austria, both addressing the automotive industry with metal oxide sensors for air quality monitoring.43 In the 2010s, micromachined metal oxide sensors targeting the air indoor quality monitoring market have been also developed. Other companies such as Figaro Engineering in Japan, the pioneer in the ﬁeld of metal oxide, start-up Cambridge CMOS Sensor (CCS) in United Kingdom, Sensirion AG in Switzerland, and large companies Bosch Sensortec in Germany and AMS in Austria are now proposing MEMS-based metal oxide sensor products. AMS has acquired AppliedSensor GmbH and Cambride CMOS Sensor to increase its technology portfolio. AMS, Bosch, and Sensirion are proposing environmental sensing solutions made of a variety of sensors, combining metal oxide sensors with temperature, humidity, pressure, optical CO2 sensors, and particle sensors, among others. Figaro Engineering Inc. investigated the potential commercialization of micromachined metal oxide gas sensors.99 The device is based on a suspended membrane etched from the front for minimizing the power consumption. They dispensed different metal oxide materials that were annealed with the integrated heater on the chip. The layer thicknesses were between below 1 mm to about 50 mm, depending on the gas to be detected. This research and development work has led to a new product, the TGS8100, for the detection of air contaminants, such as hydrogen (1e30 ppm) and ethanol, for air quality and appliance control. The sensor comes in a surface mount package with a footprint of 2.5 3.2 0.99 mm3.102 It consumes 15 mW with an applied heater voltage of 1.8V and circuit voltage of 3.0V DC pulse. It exhibits high sensitivity to cigarette smoke, cooking odors, and gaseous air contaminants with application examples such as indoor air quality monitors, air cleaners, ventilation control, and kitchen range hood control. Micromachined semiconductor gas sensors 455

Bosch Sensortec BME680 Integrated Environmental Unit is an environmental sensor for mobile applications and wearables. BME680 combines a metal oxide gas sensor for VOCs monitoring with air pressure, humidity, and ambient air temperature sensing functions within a single package. The combo MEMS solution enables multiple new capabilities for portable and mobile devices such as air quality measurement, home automation, and other applications for the Internet of Things (IoT). The sensor comes in a 3.0 3.0 mm2 footprint package with I2CandSPI communication interfaces. Applications include smart homes, smart ofﬁces and buildings, smart energy, smart transportation, HVAC, elderly care, and sport/ﬁtness. More and more devices in our surroundings are being equipped with sensors to monitor environmental parameters such as air pollution. In particular, mobile platforms such as wearables and mobile phones offer new opportunities for sensing applications. Such a combination enables for example monitoring of personal exposure to outdoor or indoor air pollutants such as NOx or volatile organic compounds that affect our health and well-being. These new applications pose a number of requirements for gas sensing technologies such as high sensitivity, good long-term stability, low power consumption, small package size, and low production costs. Sensirion’s multipixel gas sensor SGP (Sensirion Gas sensor Platform, Fig. 13.23) combines three key innovations that are crucial for the widespread integration of MOX-based gas sensors in mobile and IoT applications: long-term stability through siloxane resistance, a fully digital gas measurement solution monolithically integrated on one chip, and the integration of several sensing elements in one sensor.100 The SGP offers a complete gas sensor system integrated into compact DFN package of

Figure 13.23 The SGP multipixel gas sensor. Courtesy of Sensirion AG. 456 D. Briand and J. Courbat

On-chip Hotplate VDDH T-sensor memory

VDD

Heater System VSS controller controller Analog frontend I 2 C interface SDA Signal processing SCL

Mems Analog Digital Figure 13.24 Block diagram of the SGP multipixel gas sensor platform. Courtesy of Sensirion AG.

Sensor element: «pixel»

Read-out electrode

Heater

Figure 13.25 Micrograph of the SGP showing the four sensing elements, the readout electrodes, and the heater. Courtesy of Sensirion AG.

2.45 2.45 0.9 mm3 size. Sensirion’s CMOSens technology allows to cointegrate analog and digital electronics together with a microhotplate and the sensing elements on a single chip as shown in the block diagram in Fig. 13.24. Four MOX sensing elements based on layers of metal oxide nanoparticles are deposited on a microhotplate (Fig. 13.25). The resistance of each sensing element can be measured separately by readout electrodes. A heater and a temperature sensor are also integrated on the hotplate to actively control its operating temperature. This guarantees a stable operation of the sensor, independent of ambient temperature. The signals from Micromachined semiconductor gas sensors 457 the four sensor elements are measured by a highly optimized amplifier covering a measurement range of eight orders of magnitude. This is crucial for covering a wide variety of metal oxide sensing materials as well as different gases and gas concentrations with a single hardware platform. The signals are further processed in the digital signal processing stage with algorithms, e.g., for averaging, baseline compensation, and humidity compensation. In addition, individual calibration parameters are written during production into an on-chip memory. This allows to convert the sensor raw signals into calibrated output signals, for example concentrations of volatile organic compounds. All these features greatly simplify the integration of the SGP into different applications. The output signal can directly be used by customers as air quality indication without further processing. The combination of several MOX sensing elements on one chip brings two important advantages. First, it allows for measuring gas concentrations of several gases such as outdoor air pollutants and VOCs with one sensor. This greatly reduces cost and footprint in comparison with solutions using several sensor chips. Second, the combination of signals from different sensing elements can also be used to improve the selectivity with respect to the target gas. Traditional metal oxideebased gas sensors suffer from poor long-term stability when they are operated in atmospheres containing even very low concentrations of siloxanes, which are silicon-containing compounds found in many products of our everyday life such as cosmetics, cleaning agents, or plastic parts. The degradation caused by siloxanes typically results in a significant loss of sensitivity to VOCs and other gases as well as in a strong increase of response time.101 The degradation process and therefore the sensor life time depends on the siloxane concentration. This problem is in particular pronounced in applications like mobile phones, where the sensor is constantly exposed to high siloxane concentrations degassing from various components of the mobile phone. The core technology of the SGPdMOXSensdprovides the sensor with a unique robustness against contamination by siloxanes. This is achieved by a combination of optimization of the sensing material, operation mode, and the combination of signals from different sensing elements. The siloxane resistance significantly improves the long-term stability and accuracy of the SGP. The SGP offers a unique combination of integration, multipixel platform, and long-term stability that not only leverages MOX-based gas sensing into a new area but also opens up completely new gas sensing applications like mobile phones, wearables, and IoT devices. 458 D. Briand and J. Courbat

With respect to MEMS-based MOX gas sensors, the recent years have shown a clear trend toward utilizing sensors in the consumer space. This has led to further cost and power reduction via miniaturization and more advanced, low-cost packaging solutions, e.g., mold packages. The smaller substrate sizes gave rise to challenges with respect to deposition processes and choice of MEMS processes. The latter are nowadays more and more transferred to standard CMOS foundries and materials such as tungsten are replacing noble metals in hotplate structures. Eventually, this trend may lead to 3D-integrated or monolithic devices. A major challenge for MOX gas sensors production remains the device- to-device variation which is aggravated by the shrinking device sizes, resulting in the need to have very stable processes for both MEMS wafer manufacturing and MOX deposition for high volume production.

13.10 Conclusion Micromachined semiconductor gas sensors based on silicon microhotplate technology is now a mature technology with a few examples of devices on the market, mainly based on thick-film metal oxides (notably SnO2 and WO3). Since the end of the 1980s, the technology has evolved significantly and offers very good models for their design and robust processes for their fabrication. Various efforts have led to devices that perform very well at operational temperatures above 500 C, with homogeneous temperature distribution over the sensing area and minimum power consumption. Power consumption for continuous operation is in the order of a few mW, and sub-mW consumption can be reached using a pulsing mode of operation. These platforms can now welcome many different types of semiconducting gas sensing materials, with various formations of device array, with the very interesting possibility of modulating the operational temperature and integrating the electronics with the sensor silicon chip. The concept of microhotplates has been extended to field-effect gas sensors also with reduced power consumption and thermal cycling capabilities. Trends and perspectives are mainly in relation to nanotechnology-based devices, with the integration of nanostructured gas sensing films on conventional microhotplates and especially on polymeric-based microhotplates. New processing methods are also being investigated for the integration of metal oxide sensing layers onto microhotplate devices, such as FSP, nanowire synthesis, and the printing of metal oxide sensing layers, mainly using inkjet. Finally, fully printed version of metal oxide gas sensors has been demonstrated on large-area polymeric foil. Micromachined semiconductor gas sensors 459

References 1. Mardilovich P, Routkevitch D, Govyadinov A. Hybrid micromachining and surface microstructuring of alumina surface. In: Conf. Proc. of microfabricated systems and MEMS V, 198th ECS meeting; 2000. p. 33e42. 2. Vasiliev A, Pavelko RG, Gogish-Klushin SY, Kharitonov DY, Gogish-Klushina OS, Sokolov AV, Pisliakov AV, Samotaev NN. Alumina MEMS platform for impulse semiconductor and IR optic gas sensors. Sens Actuators B 2008;132(1):216e23. 3. Simon I, Barsan N, Bauer M, Weimar U. Micromachined metal oxide gas sensors: opportunities to improve sensor performance. Sens Actuators B 2001;73:1e26. 4. Brattain WH, Bardeen J. Surface properties of germanium. Bell Syst Tech J 1952;32:1. 5. Seiyama T, Kato A, Fujushi K, Nagatani M. A new detector for gaseous components using semiconductive thin films. Anal Chem 1962;34:1502e3. 6. Briand D, Krauss A, van der Schoot, Weimar U, Barsan N, Gopel€ W, de Rooij NF. Design and fabrication of high-temperature micro-hotplate for drop-coated gas sensors. Sens Actuators B 2000a;68:223e33. 7. Ho GW. Gas sensors with nanostructures oxide semiconductor materials. Sci Adv Mat 2011;3(2):150e68. 8. Ali SZ, Udrea F, Milne WI, Gardner JW. Tungsten based SOI microhotplates for smart gas sensors. J Microelectromech S 2008;17(6):1408e17. 9. Courbat J, Canonica M, Teyssieux D, Briand D, de Rooij NF. Design and fabrication of micro-hotplates made on a polyimide foil: electrothermal simulation and characterization to achieve power consumption in the low mW range. J Micromech Microeng 2010a;21:015014. 10. Elmi I, Zampolli S, Cardinali GC. Optimization of a wafer-level process for the fabrication of highly reproducible thin-film MOX sensors. Sens Actuators B 2008a;131: 548e55. 11. Faglia G, Comini E, Cristalli A, Sberveglieri G, Dori L. Very low power consumption micromachined CO sensors. Sens Actuators B 1999;55:140e6. 12. Graf M, Barrettino D, Zimmermann M, Hierlemann A, Baltes H, Hahn S, Barsan N, Weimar U. CMOS monolithic metal-oxide sensor system comprising a microhotplate and associated circuitry. IEEE Sens J 2004a;4(1):9e16. 13. Ç akir MC, Ç alis¸kan D, But€ un€ B, Ozbay} E. Planar indium tin oxide heater for improved thermal distribution for metal oxide micromachined gas sensor. Sensors 2016;16:1612. 14. Presmanes L, Thimont Y, el Younsi I, Chapelle A, Blanc F, Talhi C, Bonningue C, Barnabé A, Menini P, Tailhades P. Integration of P-CuO thin sputtered layers onto microsensor platforms for gas sensing. Sensors 2017;17:1409. 15. Amor S, André N, Gérard P, Ali SZ, Udrea F, Tounsi F, Mezghani B, Francis LA, Flandre D. Reliable characteristics and stabilization of on-membrane SOI MOSFET-based components heated up to 335C. Semicond Sci Technol 2017;32: 014001. 16. Cavicchi RE, Suehle JS, Kreider KG, Gaitan M, Chaparala P. Fast temperature programmed sensing for micro-hotplate gas sensors. IEEE Electron Device Lett 1995;16(6): 286e8. 17. Graf M, Gurlo A, Bârsan N, Weimar U, Hierlemann A. Microfabricated gas sensor systems with sensitive nanocrystalline metal-oxide films. J Nano Res 2006a;8(6): 823e39. 18. Barsan N, Schweizer-Berberich M, Gopel€ W. Fundamental and practical aspects in the design of nanoscaled SnO2 gas sensors: a status report. Fresnius J Anal Chem 1999; 365(4):287e304. 460 D. Briand and J. Courbat

19. Puigcorbé J, Vogel D, Michel B, VilaA,Gracia I, Cané C, Morante JR. Thermal and mechanical analysis of micromachined gas sensors. J Micromech Microeng 2003;14: 548e56. 20. Rao LLR, Singha MK, Subramaniam KM, Jampana N, Asokan S. Molybdenum microheaters for MEMS-based gas sensor applications: fabrication, electro-thermomechanical and response characterization. IEEE Sens J 2017;17(1):22e9. 21. Demarne V, Grisel A. An integrated low-power thin-ﬁlm CO gas sensor on silicon. Sens Actuators 1988;13:301e13. 22. Iwata T, Soo WPC, Matsuda K, Takahashi K, Ishida M, Sawada K. Design, fabrication, and characterization of bridge-type micro-hotplates with an SU-8 supporting layer for a smart gas sensing system. J Micromech Microeng 2017;27:024003. 23. Bosc J-M, Guo Y, Sarihan V, Lee T. Accelerated life testing for micromachined chemical sensors. IEEE Trans Reliab 1998;47(2):135e41. 24. Briand D, Tomassone G-M, de Rooij NF. Accelerated ageing of micro-hotplates for gas sensing applications. In: Proceedings of IEEE sensors 2003 conference, Toronto, Canada; 2003. p. 1314e7. 25. Briand D, Wingbrant H, Sundgren H, van der Schoot B, Ekedahl L-G, Lundstrom€ I, de Rooij NF. Modulated operating temperature for MOSFET gas sensors: hydrogen recovery time reduction and gas discrimination. Sens Actuators B 2003b;93:276e85. 26. Briand D, Beaudoin F, Courbat J, de Rooij NF, Desplats R, Perdu P. ‘Failure analysis of micro-heating elements suspended on thin membranes ‘. Microelectron Reliab 2005; 45:1786e9. 27. Lahlalia A, Filipovic L, Selberherr S. Modeling ans simulation of novel semiconducting metal oxide gas sensors for wearable devices. IEEE Sens J 2018;18(5):1960e70. 28. Ehmann M, Ruther P, von Arx M, Paul O. Operation and short-term drift of polysilicon-heated CMOS microstructures at temperatures up to 1200K. J Micromech Microeng 2001;11(4):397e401. 29. Courbat J, Briand D, de Rooij NF. Reliability improvement of suspended platinum- based micro-heating elements. Sens Actuators A 2008;142:284e91. 30. Mele L, Santagata F, Iervolino E, Mihailovic M, Rossi T, Tran AH, Schellevis H, Creemer JF, Sarro PM. A molybdenum MEMS microhotplate for high-temperature operation. Sens Actuators A 2012;188:173e80. 31. Creemer JF, Briand D, Zandbergen HW, van der Vlist W, de Boer CR, de Rooij NF, Sarro PM. Microhotplates with TiN heaters. Sens Actuators A 2008;148:416e21. 32. Graf M, Muller€ SK, Barrettino D, Hierlemann A. Transistor heater for microhotplate- based metal-oxide microsensors. IEEE Electron Device Lett 2005;26(5):295e7. 33. Lyle RP, Hughes HG, Walters D. Micromachined silicon CO gas sensors. Electrochem Soc Proc 1997;97(5):188e98. 34. Semancik S, Cavicchi RE, Wheeler MC, Tiffany JE, Poirier GE, Walton RM, Suehle JS, Panchapakesan B, DeVoe DL. Microhotplate platforms for chemical sensor research. Sensor Actuator B 2001;77:579e91. 35. Elmi I, Zampolli S, Cozzani E, Mancarella F, Cardinali GC. Development of ultralow-power consumption MOX sensors with ppb-level VOC detection capabilities for emerging applications. Sens Actuators B 2008b;135:342e51. 36. Muller€ G, Friedberger A, Kreisl P, Ahlers S, Schulz O, Becker T. A MEMS toolkit for metal-oxide-based gas sensing systems. Thin Solid Film 2003;436:34e45. 37. Friedberger A, Kreisl P, Rose E, Muller€ G, Kuhner€ G, Wollenstein€ J, Bottner€ H. ‘Micromechanical fabrication of robust low-power metal oxide gas sensors ‘. Sens Ac- tuators B 2003;93:345e9. 38. Wollenstein€ J, Plaza JA, Cané C, Min Y, Bottner€ H, Tuller HL. A novel single chip thin ﬁlm metal oxide array. Sens Actuators B 2003;93:350e5. Micromachined semiconductor gas sensors 461

39. Huotari J, Kekkonen V, Haapalainen T, Leidinger M, Sauerwald T, Puustinen J, Liimatainen J, Lappalainen J. Pulsed laser deposition of metal oxide nanostructures for highly sensitive gas sensor applications. Sens Actuators B 2016;236:978e87. 40. Gardner JW, Pike A, de Rooij NF, Koudelka-Hep M, Clerc PA, Hierlemann A, Gopel€ W. Integrated array sensor for detecting organic solvents. Sens Actuators B 1995;26e7:135e9. 41. Heilig A, Bârsan N, Weimar U, Schweizer-Berberich M, Gardner JW, Gopel€ W. Gas identification by modulating temperatures of SnO2-based thick film sensors. Sens Ac- tuators B 1997;43:45e51. 42. Kappler J, Barsan N, Weimar U, Dieguez A, Alay JL, Romano-Rodriguez A, Morante JR, Gopel€ W. Correlation between XPS, Raman and TEM measurements and the gas sensitivity of Pt and Pd doped SnO2 based gas sensors. Fresenius J Anal Chem 1998;361(2):110e4. 43. Blaschke M, Tille T, Robertson P, Mair S, Weimar U, Ulmer H. MEMS gas-sensor array for monitoring the perceived car-cabin air quality. IEEE Sens J 2006;6(5): 1268e308. 44. Courbat J, Briand D, Yue L, Raible S, de Rooij NF. Drop-coated metal-oxide gas sensor on polyimide foil with reduced power consumption for wireless applications. Sens Actuators B 2012;161:862e8. 45. Puigcorbé J, Vila A, Cerda J, Cirera A, Gracia I, Cané C, Morante JR. Thermo- mechanical analysis of micro-drop coated gas sensors. Sens Actuators A 2002;97e8: 379e85. 46. Cerda Belmonte J, Puigcorbé J, Arbiol J, Vila A, Morante JR, Sabaté N, Gracia I, Cané C. High-temperature low-power performing micromachined suspended micro-hotplate for gas sensing applications. Sens Actuators B 2006;114:826e35. 47. Espinosa E, Ionescu R, Zampolli S, Elim I, Cardinali GC, Abad E, Leghrib R, Ramírez JL, Vilanova X, Llobet E. Drop-coated sensing layers on ultra low power hotplates for an RFID flexible tag microlab. Sens Actuators B 2010;144:462e6. 48. Vincenzi D, Butturi MA, Guidi V, Carotta MC, Martinelli G, Guarnieri V, Brida S, Margesin B, Giacomozzi F, Zen M, Pignatel GU, Vasiliev AA, Pisliakov AV. Devel- opment of a low-power thick-film gas sensor deposited by screen-printing technique onto a micromachined hotplate. Sens Actuators B 2001;77:95e9. 49. Llobet E, Ivanov P, Vilanova X, Brezmes J, Hubalek J, Malysz K, Gracia I, Cané C, Correig X. Screen-printed nanoparticles tin oxide films for high-yield sensor microsystems. Sens Actuators B 2003;96:94e104. 50. Ivanov P, Stankova M, Llobet E, Vilanova X, Brezmes J, Gracia I, Cané C, Calderer J, Correig X. Nanoparticle metal-oxide films for micro-hotplate-based gas sensor systems. IEEE Sens J 2005;5(5):798e809. 51. Viricelle J-P, Pijolat C, Riviere B, Rotureau D, Briand D, de Rooij NF. Compati- bility of screen-printing technology with micro-hotplate for gas sensor and solid oxide fuel cell development. Sens Actuators B 2006;118:263e8. 52. Sahm T, M€adler L, Gurlo A, Barsan N, Pratsinis SE, Weimar U. Flame spray synthesis of tin dioxide nanoparticles for gas sensing. Sens Actuators B 2004;98:148e53. 53. Kuhne€ S, Graf M, Tricoli A, Mayer F, Pratsinis SE, Hierlemann A. Wafer-level flame- spray-pyrolysis deposition of gas-sensitive layers on microsensors. J Micromech Microeng 2008;18:035040. 54. Panchapakesan B, Cavicchi R, Semancik S, DeVoe DL. Sensitivity, selectivity and stability of tin oxide nanostructures on large area arrays of microhotplates. Nanotechnology 2006;17(2):415e25. 55. Sears WM, Colbow K, Consadori F. General-characteristics of thermally cycled tin oxide gas sensors. Semicond Sci Technol 1989;4(5):351e9. 462 D. Briand and J. Courbat

56. Lee AP, Reedy BJ. Temperature modulation in semiconductor gas sensing. Sens Ac- tuators B 1999;60:35e42. 57. Ratton L, Kunt T, McAvoy T, Fuja T, Cavicchi R, Semancik S. A comparative study of signal processing techniques for clustering microsensor data (a first step towards an artificial nose). Sens Actuators B 1997;41:105e20. 58. Kunt TA, McAvoy TJ, Cavicchi RE, Semancik S. Optimization of temperature programmed sensing for gas identification using micro-hotplate sensors. Sens Actuators B 1998;53:24e43. 59. Llobet E, Brezmes J, Ionescu R, Vilanova X, Al-Khalifa S, Gardner JW, Barsan N, Correig X. Wavelet transform and fuzzy ARTMAP-based pattern recognition for fast gas identification using a micro-hotplate gas sensor. Sens Actuators B 2002; 83(1e3):238e44. 60. Nakata S, Okunishi H, Nakashima Y. Distinction of gases with a semiconductor sensor under a cyclic temperature modulation with second-harmonic heating. Sens Actuators B 2006;119:556e61. 61. Raible S, Briand D, Kappler J, de Rooij NF. Wafer level packaging of micromachined gas sensors. IEEE Sens J 2006;6(5):1232e5. 62. Briand D, Guillot L, Raible S, Kappler J, de Rooij NF. Highly integrated wafer level packaged MOX gas sensors. In: Proceedings of the Transducers’07 conference, Lyon, France, June 10e14, 2007; 2007. p. 2401e4. 63. Suehle JS, Cavicchi RE, Gaitan M, Semancik S. Tin oxide gas sensor fabricated using CMOS micro-hotplates and in-situ processing. IEEE Electron Device Lett 1993;14: 118e20. 64. Laconte J, Dupont C, Flandre D, Raskin J-P. SOI CMOS compatible low power microheater optimization for the fabrication of smart gas sensors. IEEE Sens J 2004; 4(5):670e80. 65. Gardner JW, Guha PK, Udrea F, Covington J. CMOS interfacing for integrated gas sensors: a review. IEEE Sens J 2010;10(12):1833e48. 66. Afridi MY, Suehle JS, Zaghoul ME, Berning DW, Hefner AR, Cavicchi RE, Semancik S, Montgomery CB, Taylor CJ. A monolithic CMOS Microhotplate- based gas sensor system. IEEE Sens J 2002;2(6):644e55. 67. Barrettino D, Graf M, Taschini S, Hafizovic S, Hagleitner C, Hierlemann A. CMOS monolithic metal-oxide gas sensor microsystems. IEEE Sens J 2006;6(2):276e86. 68. Lundstrom€ I, Shivaraman S, Svensson C, Lundkvist L. A hydrogen-sensitive MOS field-effect transistor. Appl Phys Lett 1975;26:55e7. 69. Lundstrom€ I, Sundgren H, Winquist F, Eriksson M, Krantz-Rulcker€ C, Lloyd- Spetz A. Twenty-five years of field effect gas sensor research in Linkoping.€ Sens Actu- ators B 2007;121(1):247e62. 70. Kreisl P, Helwig A, Muller€ G, Obermeier E, Sotier S. Detection of hydrocarbon species using silicon MOS field-effect transistors operated in a non-stationary temperature-pulse mode. Sens Actuators B 2005;106:442e9. 71. Lloyd-Spetz A, Baranzahi A, Tobias P, Lundstrom€ I. High temperature sensors based on metal-insulator-silicon carbide devices. Phys Status Solidi 1997;162(1):493e511. 72. Briand D, Sundgren H, van der Schoot B, Lundstrom€ I, de Rooij NF. Thermally isolated MOSFET for gas sensing application. IEEE Electron Device Lett 2001;22(1):11e3. 73. Briand D, van der Schoot B, de Rooij NF, Sundgren H, Lundstrom€ I. A low-power micromachined MOSFET gas sensor. J Microelectromech S 2000b;9(3):303e8. 74. Meier DC, Semancik S, Button B, Strelcov E, Kolmakov A. Coupling nanowire chemiresistors with MEMS microhotplate gas sensing platforms. Appl Phys Lett 2007;91:063118. Micromachined semiconductor gas sensors 463

75. Li M, Yan W, Zhu H, Guo Z, Tang Z. Fabrication and characterization of a low power consumption ethanol gas sensor based on a suspended micro-hotplate. RSC Adv 2015;5:51953e60. 76. Santra S, Sinha AK, De Luca A, Ali SZ, Udrea F, Guha PK, Ray SK, Gardner JW. Mask-less deposition of Au-SnO2 nanocomposites on CMOS MEMS platform for ethanol detection. Nanotechnology 2016;27:125502. 77. Ryu K, Zhang D, Zhou C. High-performance metal oxide nanowire chemical sensors with integrated micromachined hotplates. Appl Phys Lett 2008;92:093111. 78. Marasso SL, Tommasi A, Perrone D, Cocuzza M, Mosca R, Villani M, Zappettini A, Calestani D. A new method to integrate ZnO nano-tetrapods on MEMS microhotplates for large scale gas sensor production. Nanotechnology 2016;27:385503. 79. Shao F, Fan JD, Hernandez-Ramírez F, Fabrega C, Andreu T, Cabot A, Prades JD, Lopez N, Udrea F, De Luca A, Ali SZ, Morante JR. NH3 sensing with self- assembled ZnO-nanowire mHP sensors in isothermal and temperature-pulsed mode. Sens Actuators B 2016;226:110e7. 80. Lee J, Kim J, Im JP, Lim SY, Kwon JY, Lee S-M, Moon SE. MEMS-based NO2 gas sensor using ZnO Nano-rods for low-power IoT application. J Korean Phys Soc 2017; 70(10):924e8. 81. Chen Y, Xu P, Xu T, Zheng D, Li X. ZnO-nanowire size effect induced ultra-high sensing response to ppb-level H2S. Sens Actuators B 2017;240:264e72. 82. Takacs M, ZamboD,De ak A, Pap AE, Ducs€ oC.WO€ 3 nano-rods sensitized with noble metal nano-particles for H2S sensing in the ppb range. Mater Res Bull 2016;84: 480e5. 83. Krainer J, Deluca M, Lackner E, Sosada F, Wimmer-Teubenbacher R, Koeck A, Gspan C, Rohracher K, Wachmann E, Schrems M. CMOS integrated tungsten oxide nanowire networks for ppb-level hydrogen sulfide sensing. In: Proceedings of the IEEE sensors conference, Orlando, USA, October 30 e November 3, 2016; 2016. p. 1e3. 84. Vallejos S, Umek P, Stoycheva T, Annanouch F, Llobet E, Correig X, De Marco P, Bittencourt C, Blackman C. Single-step deposition of Au- and Pt-nanoparticle-functionalized tungsten oxide nanoneedles synthetized via aerosol-assisted CVD, and used for fabrication of selective gas microsensor arrays. Adv Funct Mater 2016;23:1313e22. 85. Annanouch FE, Haddi Z, Vallejos S, Umek P, Guttmann P, Bittencourt C, Llobet E. Aerosol-asssited CVD-grown WO3 nanoneedles decorated with copper oxide nanoparticles for the selective and humidity-resilient detection of H2S. ACS Appl Mater In- terfaces 2015;7:6842e51. 86. Annanouch FE, Haddi Z, Ling M, Di Maggio F, Vallejos S, Vilic T, Zhu Y, Shujah T, Umek P, Bittencourt C, Blackman C, Llobet E. Aerosol-asssited CVD-grown PdO nanoparticle-decorated tungsten oxide nanoneedles extremely sensitive and selective to hydrogen. ACS Appl Mater Interfaces 2016;8:10413e21. 87. Steinhauer S, Chapelle A, Menini P, Sowwan M. Local CuO growth on microhotplate: in situ electrical measurements and sensing applications. ACS Sens 2016;1: 503e7. 88. Prades JD, Jimenez-Diaz R, Hernandez-Ramirez F, Cirera A, Romano-Rodriguez A, Morante JR. Harnessing self-heating in nanowires for energy efficient, fully autonomous and ultra-fast gas sensors. Sens Actuators B 2010;144:1e5. 89. McAlpine MC, Ahmad H, Wang D, Heath JR. Highly ordered nanowire arrays on plastic substrates for ultrasensitive flexible chemical sensors. Nat Mater 2007;6:379e84. 90. Briand D, Colin S, Courbat J, Raible S, Kappler J, de Rooij NF. Integration of MOX gas sensors on polyimide hotplates. Sens Actuators B 2008;130:430e5. 91. Courbat J, Briand D, de Rooij NF. Reliability of micro-hotplates on polyimide foil. In: Proceedings Transducers’11 conference, Beijing, China; 2011. p. 338e41. 464 D. Briand and J. Courbat

92. Courbat J, Briand D, de Rooij NF. Foil level packaging of a chemical gas sensor. J Micromech Microeng 2010b;20:055026. 93. Zappa D, Briand D, Comini E, Courbat J, de Rooij NF, Sberveglieri G. Zinc oxide nanowires deposited on polymeric hotplates for low-power gas sensors. In: Proceedings of the eurosensors 2012 conference, Cracow, Poland; 2012. 94. Shen W. Properties of SnO2 based gas-sensing thin ﬁlms prepared by ink-jet printing. Sens Actuators B 2012;166e7:110e6. 95. Peter C, Kneer J, Wollenstein€ J. Inkjet printing of titanium doped chromium oxide for gas sensing application. Sens Lett 2011;9:1e5. 96. Kukkola J, Jansson E, Popov A, Lappalainen J, M€aklin J, Halonen N, Toth G, Shchukarev A, Mikkola J-P, Jantunen H, Kordas K, Hast J, Hassinen T, Sunnari A, Jokinen K, Haverinen H, Sliz R, Jabbour G, Fabritius T, Myllyl€a R, Vasiliev A, Zaretskiy N. Novel printed nanostructured gas sensors. Procedia Eng 2011;25:896e9. 97. Ramírez JL, Fatima EA, Llobet E, Briand D. Architecture for the efﬁcient manufacturing by printing of heated, planar, resistive transducers on polymeric foil for gas sensing. Sens Actuators B 2018;258:952e60. 98. Rieu M, Camara M, Tournier G, Viricelle J-P, Pijolat C, de Rooij NF, Briand D. Fully inkjet printed SnO2 gas sensor on plastic substrate. Sens Actuators B 2016;236: 1091e7. 99. Ishibashi N, Kaneyasu K. Development and application of semiconductor gas sensor using MEMS technology. In: Presented at 3rd GOSPEL workshop: gas sensors based on semiconducting metal oxide e new directions, T€ubingen, Germany, November 30eDecember 1, 2009; 2009. 100. Ruffer€ D, Hoehne F, Buhler€ J. New digital metal-oxide (MOx) sensor platform. Sen- sors 2018;18:1052. 101. Ehrhardt J-J, Colin L, Jamois D. Poisoning of platinum surfaces by hexamethyldisilox- ane (HMDS): application to catalytic methane sensors. Sensor Actuator B Chem 1997; 40:117e24. 102. Khan S, Briand D. All-printed low-power metal oxide gas sensors on polymeric substrates. Flexible and Printed Electronics 2019;4:015002. CHAPTER FOURTEEN

Integrated CMOS-based sensors for gas and odor detection

P.K. Guha1, S. Santra1, J.W. Gardner2 1Indian Institute of Technology, Kharagpur, West Bengal, India 2University of Warwick, Coventry, United Kingdom

Contents

14.1 Introduction 465 14.2 Microresistive complementary metal oxide semiconductor gas sensors 467 14.3 Microcalorimetric complementary metal oxide semiconductor gas sensor 469 14.4 Sensing materials and their deposition on complementary metal oxide 472 semiconductor gas sensors 14.5 Interface circuitry and its integration 475 14.6 Integrated multisensor and sensor array systems 480 14.7 Conclusion and future trends 483 Useful web addresses 485 References 486

14.1 Introduction Gas sensors are increasingly becoming an important part of our everyday lives. They can be found in our homes (e.g., monitoring the level of CO in air from gas-ﬁred boilers), in our workplace (e.g., checking the levels of toxic gases and odors in ofﬁces), and in hospitals (e.g., monitoring anesthetic and respiratory gases during operations). There has been an increasing demand for improved workplace safety for certain industries (e.g., working in coal mines) through tougher government legislationd even in developing countries such as China and India. Moreover, there are also some emerging niche markets (e.g., sensors for notebook computers, tablets, and even mobile phones), which require very low-cost (<$5) and low-power (milliwatt or less) gas/odor sensors in high volumes (tens of millions of units per year). However, the average cost of a gas sensor available today is in excess of $25 and well over $70 for optical ones (for methane and CO2). This is because the production of discrete gas sensors tends to be

Semiconductor Gas Sensors, Second Edition ISBN: 978-0-08-102559-8 © 2020 Elsevier Ltd. https://doi.org/10.1016/B978-0-08-102559-8.00014-8 All rights reserved. 465 j 466 P.K. Guha et al. semiautomated and in batches of relatively small volumes (hundreds to thousands rather than millions). In addition, these sensors require associated interface electronics using discrete components to monitor the sensor output and drive heaters. This increases the cost even further of the sensor module with packaging to often in excess of $100. According to a recent market research report, the gas sensor market will reach around $1.3 billion by 2023, at a compound annual growth rate of 6.8% between 2017 and 2023 (“Gas Sensors Market e Global Forecast 2023” by Markets and Markets). The major factors driving the growth are due to the continuous development of miniaturized sensors, increasing enforcement of health and safety regulations by government, and increasing awareness of indoor/outdoor air quality control among users. Further growth will be generated for semiconductor-based gas sensors (as opposed to electrochemical or optical sensors) if solid-state gas sensors can be developed on a complementary metal oxide semiconductor (CMOS) chip combined with the associated circuitry using standard microelectronic and microelectromechanical processes at a reasonable cost. Success in this endeavor must reduce the volume cost of sensors to below $1e2 and, hence, access new high volume markets. Current silicon-based gas sensors can cost $50 each in low volumes, which is unacceptable to the mainstream consumer market. One of the most common and successful types of gas sensor sold is the electrochemical gas sensor that operates at ambient temperatures; however, although these are ultra-low power, they are bulky (w1cm3) and it is not possible to integrate this type of sensor and the necessary circuits on a silicon chip. This is because they require a significant volume of liquid electrolytes to operate for a sufficient lifetime and employ non-CMOS catalytic materials, such as platinum or silver electrodes. Another successful gas sensor is based on nondispersive infrared absorption, but low-cost versions have poor limits of detection (w100s ppm), require high power for the source, and require expensive non-CMOS optical band-pass filters. They are also even more costly than electrochemical sensors as stated above at $75 or more with electronics and packaging. In this chapter, we discuss the two main types of low-cost semiconductor gas microsensors: (1) resistive and (2) calorimetric. Recent research reports that it is possible to integrate these types of gas sensor onto a single CMOS die and, hence, they have the potential for costs low enough to open up mass markets in the near future. Here we address the key issues associated with the integration of these gas sensors onto a single CMOS die. The Integrated CMOS-based sensors for gas and odor detection 467 main challenges are to deposit/grow gas-sensitive materials (CMOS, in particular, imposes many constraints on material integration); isolate the sensing area from interface electronics present in the same silicon die; and isolate the nature of the electronic circuits required to interface the sensors and power the microheater. We also briefly discuss the potential for low- cost arrays of gas sensors, such as those required for complex odor detection, namely electronic nose (e-nose) applications. Finally, we consider the future of smart CMOS gas sensors with regard to low-cost, low-power devices needed for the emerging Internet of things (IoT).

14.2 Microresistive complementary metal oxide semiconductor gas sensors Resistive gas sensors are based on monitoring a change in the electrical conductivity of a sensing material (primarily, nonstoichiometric n-type and p- type semiconducting metal oxides, such as SnO2,WO3, ZnO, TiO2, CuO, and NiO) at high operating temperatures compared with standard CMOS technology (<þ125 C) (typically, 200e500 C) in the presence of the hazardous/odorous gas. These are perhaps one of the earliest types of available gas sensor on the market. For example, Taguchi gas sensors (TGS) were commercialized in the early 1970s and are still produced today by Figaro Inc., Japan. The TGS comprises a mm-sized platinum heater coil inside a ceramic tube several mm long. A thick layer of porous tin oxide ﬁlm is attached by hand over thick platinum electrodes on the ceramic cylinder and sintered at a high temperature (e.g., 650 C) to achieve the appropriate crystalline grain structure. Fig. 14.1 shows the structure of a Taguchi-type

Lead wire Electrode Lead wire

Ceramic tube

Heater coil Sintered tin oxide Figure 14.1 Basic structure of a Taguchi-type tin oxide gas sensor (Figaro Engineering, Japan). 468 P.K. Guha et al. tin oxide gas sensor. The electrical resistance of the tin oxide coating changes in the presence of any gas that reacts with chemisorbed oxygen species (O , O2 , etc) at high temperature (typically 250e450 C), and this is measured via the electrodes and a basic potential divider circuit with a matched reference resistor. The reasons for the high-temperature operation of metal oxide gas sensors are (1) the reaction with chemisorbed oxygen sites (and catalytic dopants) changes and can be more specific at higher temperatures, (2) the reaction kinetics are generally much faster, although the sensitivity may decrease too, and (3) the interference from water vapor is reduced above 150 C.1 Platinum wire resistive heaters are used to heat up the gas sensing material to high operating temperatures for the reaction to be efficient but require several hundred milliwatts of power. The key challenges associated with resistive CMOS sensor design are to reduce the size and power consumption significantly and to achieve a stable and uniform temperature over the sensing area. New-generation microresistive sensors are usually formed on a thin thermally insulated silicon oxide or nitride membrane so as to reduce the power loss; this thin membrane-like structure is often referred to as a “silicon microhotplate.” Early researchers (before w2005) studied different materials for designing the electrically conducting microheater structures, such as e doped polysilicon, doped single crystal silicon, aluminum, and platinum.2 4. Platinum heaters are probably the most stable material at 400 C, but this material is not available in a CMOS foundry; hence, post-CMOS deposition is necessary. Within the CMOS process, polysilicon has a poor long-term stability at high temperature because of electromigration, grain boundary movement, and crack propagation, while aluminum shows electromigration at high temperatures (aluminum has a melting point of 660 C, prolong heating over 350 C softens the material and can cause electromigration). Metal oxide semiconductor field-effect transistor heaters can be controlled by applying varying bias voltage at the gate; however, very high temperature operation is not reliable in silicon on insulator (SOI) CMOS because of the triggering of a parasitic bipolar transistor.5 In the literature, many researchers have claimed that their design is “CMOS compatible,” where in most of the cases “CMOS compatible” means that either the sensor design can be made using a modified CMOS process or CMOS materials have been used in the fabrication. However, no standard CMOS process has been used to produce their respective devices or components.6 There have been very few groups worldwide who actually used commercial CMOS-MEMS foundry to fabricate sensor/circuit Integrated CMOS-based sensors for gas and odor detection 469 single chip solution, e.g., Prof. Gardner’s group at Warwick University, Prof. Udrea’s group at Cambridge University, Prof. Henry Baltes’s group at ETH Zurich, Dr. Semancik, and others at NIST to name a few. þ þ Microheaters made from p silicon perform better than n silicon and much better than doped polysilicon in long-term operation (using industrial CMOS foundry).7 Some commercial foundries now support high- temperature metals (after 2005) along with aluminum metallization in the same wafer; so, high-temperature metal (e.g., tungsten) can be used for the heater and aluminum (lower resistivity compared with tungsten) for circuit metallization. Work led by a group from Cambridge University using SOI CMOS wafer reported the superiority and reliability of a tungsten heater.8 Here, the heater is embedded in an oxide/nitride membrane, hence, thermally isolated from the rest of the chip, which ensures reliable on- chip circuit operation. The membranes were formed at the wafer level (from MEMS foundry) using a backside deep reactive ion etch (i.e., a post-CMOS MEMS process) of the handle silicon. The heaters can heat up the membrane and to, say, 300 C only require 6 mW of power. The interdigitated electrode (for resistance measurement of the sensing material), formed by the top metal layer, was exposed at the same step of bond pad opening. The cross-sectional view of the sensor and a top view of a fabricated device are shown in Fig. 14.2(a) and (b). Other research groupsdsuch as Baltes (ETH, Zurich) and de Rooij (IMT, Neuchatel)dhave reported equally power-efficient non-SOI microheaters fabricated inside silicon plugs.9,10 Microhotplates can also be made using front etch of silicon, where the membrane is suspended via bridges.11 Such microhotplates are more power efficient (because the hotplate is surrounded by air) compared to back- etched one but may be less stable.

14.3 Microcalorimetric complementary metal oxide semiconductor gas sensor Microcalorimetric sensors detect heat or the change in enthalpy generated from the chemical reaction that takes place between the sensing material and the gas molecules. This kind of sensor generally functions at higher temperatures than resistive gas sensors (around 500 C or more), and there is a need to integrate the microheater under the catalytic sensing area, as in the case of microresistors. One of the popular calorimeters is called the “pellet resistor” or “pellistor,” which was designed and patented by English Electric Valve (EEV) Ltd in the early 1970s (www.e2v.com). 470 P.K. Guha et al.

(a) Metal micro-heater Interdigitated Sensing layer electrode Silicon-dioxide Silicon nitride

Buried oxide Handle silicon

Handle silicon etched by DRIE

Gas sensor area Circuit area (b)

Figure 14.2 (a) Cross-sectional view of complementary metal oxide semiconductor microhotplate-based gas sensor (drawing not to scale). (b) An optical microscope image of the fabricated microhotplate device. Adapted from Santra S, Ali SZ, Guha PK, Covington JA, Zhong G, Robertson J, Milne WI, Gardner JW, Udrea F. Post-CMOS wafer level growth of carbon nanotubes for low-cost microsensorsda proof of concept. Nanotechnol 2010a;21:485301. 7pp.

Conventional pellistors (shown in Fig. 14.3) are made of a platinum coil in a porous alumina bead (diameter of a few millimeters) along with an embedded metal catalyst (e.g., palladium). The platinum coil not only heats up the bead to enable the reaction but also acts as a resistance thermometer to measure the change in temperature of the catalytic surface relative to a Integrated CMOS-based sensors for gas and odor detection 471

Figure 14.3 A conventional commercial pellistor (City Technology Ltd, UK, now part of Honeywell). second noncatalytic reference beaddin a potential divider circuit. The latest designs of calorimeters are called microcalorimeters and usually use membrane technology such as microresistive sensors (except these do not need the sensing electrodes). Two identical sensors (one having an active sensing material and the other playing the role of a reference) and a differential ampliﬁer are often used to measure the difference in temperature caused by the chemical reaction, as shown in Fig. 14.4. Two accurate temperature

Figure 14.4 A differential transducing circuit for microcalorimeter. Adapted from Udrea F, Gardner JW, Setiadi D, Covington JA, Dogaru T, Lu CC, Milne WI. Design and simulations of SOI CMOS micro-hotplate gas sensors. Sensor Actuator B 2001;78:180e90. 472 P.K. Guha et al. sensors placed within the membranes are used to detect the temperature in the respective microcalorimetric sensors.4

14.4 Sensing materials and their deposition on complementary metal oxide semiconductor gas sensors Modern solid-state (resistive) gas sensors are generally based on wide- band-gap semiconducting metal oxideesensing materials. Resistive gas sensors available in the market (e.g., TGS 812, Figaro Engineering) generally use doped tin oxide as the sensing material. Solid thin film metal oxides are not effective for microsensors (the smaller surface area causes lower sensitivity). Thus, nanomaterials are suitable candidates for gas detection because of their exceptionally high surface area and dimension compared with the depletion region. So, there is a possible increase in sensitivity even for a miniaturized sensor area using new nanomaterials with even smaller grain sizes or dimensions and hence greater total surface areas. Metal oxideebased sensing layers are popular because they give reasonably good response toward different gases and volatile organic compounds (VOCs). However, they have two major drawbacks: (1) metal oxides require elevated temperature to achieve good sensitivity (hence consume large power) and also (2) they respond to more than one gas and VOC (hence have poor selectivity). The first issue can be sorted out by using more power-efficient microhotplates. Fast microhotplates (because of their low thermal mass) can be used in a pulse mode to reduce power consumption even further. In the last few years, there have been reports of two-dimensional (2D) materials as sensing layer, e.g., reduced graphene oxide,12 molybdenum disulphide 13 14 (MoS2), tungsten disulphide (WS2), etc. Such materials can interact with chemical analytes at near room temperature. Thus, they are promising candidate materials for very low-power resistive sensors. However, the sensitivity of these materials is usually poor compared with high- temperature n-type metal oxides. This can be improved if 2D layered materials are mixed with metal oxides (to get the advantage of both), i.e., large response at near room temperature.15 The second issue can be improved by various techniques, e.g., (1) functionalizing the sensing layer by metal nanoparticles,16 (2) using mixed metal oxides,17 or (3) analyzing the sensor array data through pattern recognition technique (e.g., principle component analysis, artificial neural network). Integrated CMOS-based sensors for gas and odor detection 473

However, growing nanomaterials directly on CMOS wafer is very challenging; the CMOS substrate cannot sustain prolonged high-temperature heating (because high temperature can cause rediffusion in active area of CMOS transistor and also deformation in metal tracks because of aluminum electromigration), which is often necessary during the growth process and annealing of the nanomaterials. Again, it is recommended not to use any harsh chemicals and environments (e.g., long plasma exposure might damage fragile microhotplate structure) for nanomaterial synthesis on CMOS. Therefore, a low temperature growth process (<500 C) with CMOS- friendly chemicals is highly desirable for sensing material synthesis. One such method has been reported for growing zinc oxide nanowires on a fully processed CMOS microhotplate.18 A scanning electron microscopy picture of grown nanowires on an SOI device is shown in Fig. 14.5. Recently, a local growth technique has also been reported for carbon nanotubes (CNTs)dagain, on microhotplate area.19 Here, microheaters are used to form catalytic islands and, hence, CNTs are grown only on the heater areas without exposing the whole CMOS chip to problematic high temperatures. This idea has been extended to grow CNTs at the wafer level in speciﬁed areas of multiple chips simultaneously (scheme shown in Fig. 14.6), which later can be diced and thus serve the purpose of post-CMOS batch fabrication of sensing material.

(a)

(b)

Figure 14.5 (a) Scanning electron microscopy image of the zinc oxide nanowires on a complementary metal oxide semiconductor microhotplate. (b) Higher magniﬁcation view of nanowires. Adapted from Santra S, Guha PK, Ali SZ, Hiralal P, Unalan HE, Covington JA, Amaratunga GAJ, Milne WI, Gardner JW, Udrea F. ZnO nanowires grown on SOI CMOS substrate for ethanol sensing. Sensor Actuator B 2010b;146:559e65. 474 P.K. Guha et al.

Figure 14.6 Scanning electron microscopy image of carbon nanotubes, which were locally grown simultaneously using microheaters placed at adjacent chips. Adapted from Santra S, Ali SZ, Guha PK, Covington JA, Zhong G, Robertson J, Milne WI, Gardner JW, Udrea F. Post-CMOS wafer level growth of carbon nanotubes for low-cost microsensorsda proof of concept. Nanotechnol 2010a;21:485301. 7pp.

There are many possible ways of depositing nanomaterials onto a CMOS substratedfor example, chemical vapor deposition (CVD) is possible provided the process temperature can be reduced to below 600 C. One type of CVD is aerosol-assisted CVD; in this process, precursors are transported to the substrate by means of a liquid/gas aerosol, which can be generated ultrasonically. This technique is suitable for use with nonvolatile precursors. Alternatively, very thin layers of metal oxides can be grown by using an atomic layer deposition process. Another promising recent approach is flame spray pyrolysis because it enables the control of the size of the metal oxide nanocrystals in a high-temperature flame before deposition on a wafer close to room temperature. However, the above-mentioned high-temperature synthesis techniques might not always be effective to realize highly sensitive sensing layer. This is because a high-temperature synthesis technique can reduce the number of active defect sites in the nanostructure. Thus, simple low-cost chemical routes (e.g., modified hummers method for graphene oxide synthesis, liquid exfoliation for 2D material synthesis, and hydrothermal techniques for metal oxide synthesis) might be a better approach (simpler and lower cost) to get nanostructures with ample defect sites. Techniques such as spray coating and even inkjet printing can be very useful for depositing chemically grown nanomaterials on specified areas; particularly, printers with high precision are already available in the market Integrated CMOS-based sensors for gas and odor detection 475

(Sonoplot, Tesscorn). The advantage of this approach is that one can avoid several steps of conventional lithography and that use of very high temperatures and harsh chemicals is not required. However, sometimes, nanomaterials are not easy to dissolve or form stable suspensions in a liquid; in this case, it may be necessary to use a high-speed sonicator to avoid the nozzle clogging up in the printer or spray-coater. Also, use of properly viscous nanomaterials reduces the chances of nozzle clogging. Again, in situ growth of nanomaterial directly on the device gives better adhesion compared to spray coating/inkjet printing. Therefore, in latter cases, one might need to carry out surface treatment (e.g., roughening of device surface through mild oxygen plasma) before sensing layer deposition.

14.5 Interface circuitry and its integration There are two possible ways of interfacing gas sensors with their associated interface circuitry: the hybrid approach and the monolithic approach: • Hybrid approach: this method employs separate chips for the sensors and the circuits. So, one can reuse the same circuit chip, even if there is a problem with the sensor device (e.g., breaking of the membrane), and enhance manufacturing yield through screening. The other main advantage is that there is no restriction on the type of material used and the process steps for the fabrication of the sensors; hence, both the material and the fabrication process can be adjusted to optimize the overall performance of the sensor devices. However, parasitic capacitances or inductances associated with long interconnect and bonding wires are undesirable and can cause higher levels of signal degradation/noise. In addition, the hybrid approach is more expensive than a single-chip implementation when considering high volumes. Sometimes this approach is called a system in a package. • Monolithic approach: in this method, the sensor and the circuits are on the same silicon chip. The monolithic approach is cost-effective in high unit volumes, making it commercially attractive. From the performance point of view, this system offers advantages such as a signiﬁcant reduction in the noise and cross-talk levels as the number (and length) of wires and bonds is substantially reduced and there should be much better matching between different sensors. However, a fault in a sensor will result in the failure of the complete chip, even if the circuitry is working properly. This approach is also called a system on a chip. 476 P.K. Guha et al.

Apart from the above two methods of integration, there has been recent development to integrate sensors on top of CMOS substrate, namely three- dimensional (3D) integration. In this approach, different building blocks, e.g., analog front end, digital electronics, memory, wireless module, and multisensor arrays, can be fabricated on separate silicon substrates from same/different foundries and integrated through-silicon vias to achieve 3D stacking. This will reduce delay (because of reduction in long metal tracks present in a complex monolithic chip), add additional functionality, increase ﬂexibility (because of possibility of fabrication from different foundries), and of course signiﬁcantly reduce silicon footprint. However, it is extremely challenging to integrate so many functionalities, proper thermal management, and package of entire system. A range of methods of 3D stacking have been explored recently, starting from monolithic stacking to die bonding and wafer bonding, to name a few. A common problem with any microsensor is that they tend to generate small electrical signals at the sensor output, so the front-end circuitry needs to be of very low noise (i.e., a good analog design is required). Present submicron CMOS processes offer very low rail-to-rail supply (say, 0e1.8 V); this is good for low power but makes front-end design much harder because of sensor drift (drift occurs because of aging and incomplete release of analyte at the end of a measurement, etc.). As chemoresistive sensors are based on a change in conductivity at high operating temperatures, so main circuit blocks required for interfacing these sensors are driving circuits for microheaters, temperature control units, and sensing material interface circuits. • Driving circuits for microheaters: this generally requires a stable, accurate, controllable current or voltage source circuit with large current (tens of milliamps) driving capability. • Temperature control unit: A bell-shaped temperature dependence of the gas sensor response is most commonly observed experimentally with a distinct maximum depending on the sensing material used and corresponding gas analytes; thus the temperature of the gas sensing material plays a vital role in improving the selectivity of the sensor. Hence, accurate control of the temperature of the gas sensing material is extremely important. In most commercial gas sensors, the change in heater resistance with temperature is used to set or control the operating temperature of the heater itself. In one such reported work, Bota et al.20 usedthesameresistorelementastheheaterandtemperaturesensor. They used a pulse width modulation technique where, during the off Integrated CMOS-based sensors for gas and odor detection 477

mode of the heater, a small current was driven through the heater to measure its temperature (as shown in Fig. 14.7). However, a separate temperature sensor (in the form of silicon/polysilicon resistor or silicon diode) can increase the ﬂexibility in circuit design and, hence, the reading and control of the precise temperature of the sensing area. There are several ways to control the heater temperaturedfor example, the heating can be controlled in oneoff mode (sometimes called a “bangebang” controller), proportional mode, or proportionaleintegralederivative (PID) mode. The oneoff controller is the simplest form: if the microhotplate is cooler than a setpoint temperature, the heater is turned on at maximum power; once it is hotter than the setpoint temperature, the heater is switched off completely. However, bangebang control can give rise to instability unless controlled carefully. It might also damage the heater because of very rapid changes in voltage (hence, temperature). In this respect, on- chip digital PID controller is a much better option.21 It usually has rapid respond time without overswing and steady state error; hence, smooth temperature control of the heater is possible.

Figure 14.7 Diagram of a temperature control circuit, in which the same heater is used for measuring the temperature. Adapted from Bota SA, Dieguez A, Merino JL, Casanova R, Samitier J, Cane C. A monolithic interface circuit for gas sensor arrays: control and measurement. Analog Integr Circuits Signal Process 2004;40:175e84. (VDD, supply voltage. Vref, reference voltage). 478 P.K. Guha et al.

• Sensing material interface circuit: the design of this interfacing circuit is one of the main challenging components of a resistive gas sensor. This is because of the following: (a) The circuit must handle the precision and dynamic range that is demanded by the gas sensing element. The baseline resistance varies enormously depending on the sensing material used, e.g., sensing materials such as metal oxides can have sheet resistances from kU per square up to even GU per square. (b) Often, there is drift in the baseline resistance of the sensing material; this can be due to a variety of reasons (e.g., the material is not thermodynamically stable, material has been polluted by the targeted chemical analytes). The circuits should be able to compensate for this drift to give accurate changes in resistance in the presence of the target gas. The literature reports many different solutions to the challenges mentioned brieﬂy above. A very simple scheme is to use a resistor divider circuit or a Wheatstone bridge technique. But these are not ideal when seeking full integration, as they need either trimming or variable resistors from outside the chip to match the sensor resistor. Also, simple voltage divider will require a resistor bank circuit along with very large value of resistors to cover wide range of sensing material resistance, which will consume a large silicon area. One of the popular approaches found in the literature is the resistance-to-frequency conversion. The challenge with such a scheme is to retrieve/isolate parasitic capacitances associated with the sensing material because this is a high-frequency process and, hence, resistance measurement will be contaminated with sensing material capacitance. Two such schemes have been reported by De Marcellis et al.22 and Grassi et al.23,24 In the scheme reported by De Marcellis et al., the generated pulse provides both sensing resistance values and parasitic capacitances associated with the sensing material. In the scheme reported by Grassi et al., (shown in Fig. 14.8), the parasitic capacitor role can be avoided by isolating the sensor resistor from the oscillator portion of the circuit. This was achieved by using a sensing material resistor at the reference arm of a current mirror and then using the mirror current to charge and discharge a capacitor. The capacitor voltage is then fed into comparators to compare with high (VH) and low voltage (VL) levels to generate a digital pulse. This resultant square wave time period has the information of the sensing material resistor. The scheme achieved (simulation result) a worst-case precision of about Integrated CMOS-based sensors for gas and odor detection 479

VDD

VH + CNTy Vref CNTx – CNTx + Q – – CNTy + CNTx + CNTy Sensing material – VL Flipflop

Figure 14.8 Resistance controlled oscillator circuit for gas sensing material interface. Q and Q are two complementary signals at the output of the ﬂip ﬂop. Adapted from Grassi M, Malcovati P, Baschirotto A. A 141-dB dynamic range CMOS gas-sensor interface circuit without calibration with 16-bit digital output word. IEEE J Solid State Circuits 2007;42: 1543e1554.

0.5% over the range of ﬁve decades (1 kUe100 MU), as shown in Table 14.1. A relatively simple scheme has been reported by Barrettino et al.25 A large dynamic range of sensing material was compressed by using two diodes (logarithmic converter approach); however, the scheme generated only 8- bit resolution (the circuit is shown in Fig. 14.9). One popular baseline drift removal approach has been reported by Koickal et al.26,27 and is shown in Fig. 14.10. In this scheme, during the setup phase, each sensor is driven by a small value current source and the voltage across the sensor is digitally stored using a simple counting analog-to-digital converter (ADC). This stored value is converted back to an analog signal (using a digital-to-analog converter) and then subtracted from the sensor signal, thus removing the

Table 14.1 Simulation precision results for sensors resistance materials.

Rnominal, applied Rmeasured, simulated Linear error (%) 1kU 1.0183 kU 0.30881 100 kU 102.27 kU þ0.13148 1MU 1.0212 MU 0.01593 100 MU 101.90 MU 0.04878 480 P.K. Guha et al.

Figure 14.9 Logarithmic compression circuit for gas sensing material interface. (Vcontrol is a voltage applied from outside, which will ﬁx the voltage across the sensing material and thus ﬁx the current at the reference arm of the current source. Vout1 and Vout2 are two output terminals of the circuit.) Adapted from Barrettino D, Graf M, Zimmermann M, Hierlemann A, Baltes H, Hahn S, Barsan N, Weimar U. A smart single-chip micro-hotplate- based chemical sensor system in CMOS-technology. Int Symp Circuits Syst 2002;2:157e60. baseline resistance. The scheme achieved good sensor stability, and the drift of baseline resistance was found to be less than 5 mV. More advanced signal processing circuitry is best performed using digital blocks and microcontroller unit. Digital blocks are easier to design and consumed less power. Delta-sigma ADC is often used to reduce noise of the sensor signal. For smart sensor system, low-power microcontroller interfacing is very useful. This will help in controlling sensor signal, data processing (e.g., pattern recognition operation), and also transferring sensor data to mobile platform.

14.6 Integrated multisensor and sensor array systems There is an increasing desire to measure more than one gas simultaneously. For example, in a boiler combustion process, there could be a requirement to measure hydrogen or methane, as well as CO, CO2, and even oxygen. Alternatively, in the case of automotive gases, there is a desire Integrated CMOS-based sensors for gas and odor detection 481

Figure 14.10 Diagram for baseline cancellation circuit. (Clk, clock; DAC, digital to analog converter; R1, resistance; Vout, output voltage of the circuit.) Adapted from Koickal TJ, Hamilton A, Tan SL, Covington JA, Gardner JW, Pearce TC. Analog VLSI circuit implementation of an adaptive neuromorphic olfaction chip. IEEE Trans Circuit Syst I: Regular Paper 2007;54:60e73.

to measure CO, NO2, and unburnt hydrocarbons. In other words, there is an increasing demand to have multiegas sensing systems with perhaps between four and eight sensors (including ambient humidity, ambient temperature, and even barometric pressure). In some ways, this is the same requirement faced when developing electronic nose technology. Sensor-based electronic noses are based on an array of nonspeciﬁc gas sensors coupled with a pattern recognition technique.28,29 Electronic noses tend to be large, expensive instruments. For example, the Fox 4000 electronic nose (Alpha MOS, France) comprises 18 power- hungry Taguchi-like metal oxideeresistive gas sensors (see Fig. 14.11). Consequently, there is a signiﬁcant market pull for sensor arrays containing selective gas sensing materials (multigas devices) and partially selective gas 482 P.K. Guha et al.

Figure 14.11 Fox 4000 electronic nose (Alpha MOS, France). sensing materials (electronic noses). The two main challenges in sensor arrayebased technology is cost and power consumption. First, the cost of sensor arrays will be signiﬁcant; a set of four gas sensors will probably cost $100. This, with the addition of discrete circuitry, display, packaging, etc., leads to prices of $500e$1000 and, so, a limited market take-up. Second, the power consumption of a set of six metal oxide gas sensors will be 1e2 W. This is not compatible with their implementation of battery-operated devices (such as mobile phones and tablets)dneither would it be suitable for automotive applications, where low power is still a major demand. The power consumption of SOI-based gas sensing is much lower than other technologies and enables individual devices to operate at below 10 mW DC8; through the pulsing of the heater, it is possible to obtain a power consumption of less than 1 mW per device. In addition, the use of deep reactive ion etching permits the integration of at least four microhotplates on a single 1 mm by 1 mm die. For example, Fig. 14.12 shows 1 mm square silicon dies with two and four separate microhotplates on them. The four-element chip can run four gas sensors at different temperatures with low-power consumption (power consumption of less than 25 mW at 400 C). The small thin SOI membranes provide a thermal response time of about 5 ms and so these devices can be operated in a pulsed mode and, hence, reduce the power consumption by a factor of 10 while still sampling every second. Integrated CMOS-based sensors for gas and odor detection 483

Figure 14.12 Multiple microhotplate-based gas sensors on a single silicon die (Cam- bridge CMOS Sensors, UK now ams Sensors UK Ltd).

The utilization of SOI CMOS technology enables on-chip annealing at 650 C and operating temperatures of up to 600 C, thus allowing the use of not only metal oxideeresistive materials (e.g., SnO2 or WO3) but also catalytic metals (e.g., palladium or platinum). With a CMOS or BiCMOS process comprising analog components, it is now possible to combine a set of four sensors with analog sensing/drive circuitry and a digital controller in a die of only 1 mm by 2 mm, thus permitting the production of smart gas sensors in high volumes for only $1e2. In practice, the cost of mounting and wire bonding the chip onto, say, a TO-5 package is likely to cost more than the silicon die itself!

14.7 Conclusion and future trends In this chapter, we have discussed technological developments toward the reduction in the size, power, and, most importantly, cost of both resistive and calorimetric gas (and odor) sensors. A recent CambridgeeWarwick University spin-out Cambridge CMOS Sensors Ltd in the United Kingdom (acquired by ams in June 2016 and now called ams Sensors UK Ltd) believes that it is possible to fabricate a set of gas sensors and associated circuitry on a single silicon die at a cost of $1e2 in volume production and with a power consumption of a few milliwatts. This can be achieved through the use of high-temperature tungsten interconnects in an SOI CMOS process and deep reactive ion etching on the reverse of the silicon wafers. The low 484 P.K. Guha et al. thermal mass of the SOI membrane results in a thermal response time of just a few milliseconds. This, in turn, makes the technology suitable for using more advanced signal processing methods that modulate the device temperature. For example, it has been shown that switching the operating temperature between set points and measuring the time-series signals can result in a single microsensor measuring more than one gas!30 The method is based on the principle that different gases react at different rates and so the dynamical information can be used not only to identify different gases but also simple mixtures of gases. The future of smart gas sensors may also be based on new sensing materials. SOI CMOS technology permits the on-chip deposition of new, exciting materials using CVD at 600e750 C. For example, it is now possible to grow zinc oxide nanowires and CNTs directly onto the gas sensors.18,19 Similarly, it is also possible to grow a single layer of graphene directly on the chip at 650 C with a nickel rather than copper catalyst. Although, the response of CNTs and graphene to gases has been shown recently to suffer from poor sensitivity, selectivity, and response times, it may be that synthesis via chemical route and functionalization will overcome these problems and open up a clear route to successful commercialization. Such low-power, smart, and selective chemical sensors will be appropriate in many applicationsdsafety, security, environmental monitoring, and health care to name a few. In the case of biomedical detection, a nonin- vasive breath test approach looks very promising, as it is easy to repeat the detection process and it does not cause the discomfort associated with blood tests. The breath test method is inexpensive and also point-of-care detection can be undertaken. It has been found that specific gases/vapors correspond to specific diseasesdfor example, acetone (for diabetes), nitric oxide (for asthma), carbon dioxide (for monitoring respiratory patients to evaluate their lung and pulmonary function), etc. In such health care applications, the cost of the sensor will be a major factor because one requires simple plug-and- play low-cost sensors. We envisage smart sensor modules that cost less than V5 each and are 1 mm by 1 mm and so could be packaged in a micro-SD card and inserted into tablets and even mobile phones. This means that they can be used in home automation (IoT); for example, users would be able to monitor many types of environmental gases, such as combustible CH4 and toxic CO levels from domestic boilers. More challenging will be addressing health care for an aging populationdagain, in homes. With more advanced circuitry and signal processing techniques, it Integrated CMOS-based sensors for gas and odor detection 485 should be possible to measure the breath gases mentioned above with results wired straight to doctors’ surgeries or hospitals. In last one decade, people have also been trying to develop sensors on low-cost substrate, e.g., sensor on cloth, paper, glass, and plastic. This will reduce the sensor cost drastically provided batch fabrication is possible. However, one needs to use hybrid approach, i.e., separate interface electronics chip, in this case. Temperature of operation is an inherent property associated with a particular material. In this respect, layered materials have clear advantage compared with metal oxide counterpart. Recently, there have been reports of self-power sensors. Here sensor device is interfaced with some energy harvesting unit (energy harvesting is a process by which ambient energy [e.g., solar, mechanical, and thermal] is captured and converted to electricity to drive a sensor device). It can be inconvenient to replace batteries in devices that need to work over long periods of time or in a remote place. We believe that the next generation of computer tablets, smart phones, and wearable electronic systems will contain integrated gas sensors and, thus, CMOS gas sensors will find their way into everyday use in the next few years. Also, with the rapid expansion of cloud computing and the IoT, mobile sensors are recognized as an essential component in future ubiquitous sensor networks. Thus, the sheer numbers (trillions) of future sensor devices, i.e., “things” of IoT, will negate the possibility of individual sensor mainte- nance through human intervention and frequent battery replacement. The requirement to deploy huge number of nodes of smart sensor systems will be useful at smart city or smart building for environmental monitoring. Such sensor systems need to communicate with each other and also with central sensor hub. Each sensor will have unique identity and will be connected to existing internet infrastructure. This IoT concept will be effectively realized if development of low-power low-cost sensor system is possible, and in this respect sensor integration with CMOS platform might play a vital role. Thus, huge IoT market of trillions (w1012) of connected devices will drive and speed up to realize smart sensors; this is an exciting prospect for sensor manufacturers, investors, and end users.

Useful web addresses

http://www.alpha-mos.com/ http://www.sonoplot.com 486 P.K. Guha et al.

http://www.customsensorsolutions.com http://www.e2v.com http://www.figarosensor.com/ https://www.marketsandmarkets.com/PressReleases/gas-sensor.asp http://www.tesscorn.com/nanotechnology.htm References [1] Gardner JW, Varadan VK, Awadelkarim OO. Microsensors MEMS and smart devices. West Sussex, England: Wiley; 2001. [2] Cavicchi RE, Suehle JS, Kreider KG, Shomaker BL, Small JA, Chaparala P, Gaitan M. Growth of SnO2 films on micromachined hotplates. Appl Phys Lett 1995;66:812e4. [3] Houlet LF, Tajima K, Shin W, Itoh T, Izu N, Matsubara I. Platinum micro-hotplates on thermal insulated structure for micro-thermoelectric gas sensor. IEEJ Trans Sensors Micromachines 2006;126:568e72. [4] Udrea F, Gardner JW, Setiadi D, Covington JA, Dogaru T, Lu CC, Milne WI. Design and simulations of SOI CMOS micro-hotplate gas sensors. Sensor Actuator B 2001;78: 180e90. [5] Guha PK, Ali SZ, Lee CCC, Udrea F, Milne WI, Iwaki T, Covington JA, Gardner JW. Novel design and characterisation of SOI CMOS micro-hotplates for high temperature gas sensors. Sensor Actuator B Chem 2007;127:260e6. [6] Graf M, Barrettino D, Zimmermann M, Hierlemann A, Baltes H, Hahn S, Barsan N, Weimar U. CMOS monolithic metal-oxide sensor system comprising a microhotplate and associated circuitry. IEEE Sens J 2004;4:9e16. [7] Iwaki T, Covington JA, Gardner JW, Udrea F, Blackman CS, Parkin IP. SOI-CMOS based single crystal silicon micro-heaters for gas sensors. In: Proceedings of IEEE sensors conference. Korea; 2006. [8] Ali SZ, Udrea F, Milne WI, Gardner JW. Tungsten-based SOI microhotplates for smart gas sensors. J Microelectromech Syst 2008;17:1408e17. [9] Barrettino D, Graf M, Song WH, Kirstein K, Hierlemann A, Baltes H. Hotplate-based monolithic CMOS microsystems for gas detection and material characterization for operating temperatures up to 500 C. J Solid-State Circuits 2004b;39:1202e7. [10] Briand D, Schoot BVD, Rooij NFD, Sundgren H, Lundstrom I. A low-power micromachined MOSFET gas sensor. J Microelectromech Syst 2000;9:303e7. [11] Afiridi MY, Suehle JS, Zaghloul ME, Berning DW, Hefner AR, Cavicchi RE, Semancik S, Montgomery CB, Taylor CJ. A monolithic CMOS microhotplate- based gas sensor system. IEEE Sens J 2002;2(6):644e55. [12] Ghosh R, Midya A, Santra S, Ray SK, Guha PK. Chemically reduced graphene oxide for ammonia detection at room temperature. ACS Appl Mater Interfaces 2013;5(15): 7599e603. [13] Burman D, Ghosh R, Santra S, Ray SK, Guha PK. Role of vacancy sites and UV- ozone treatment on few layered MoS2 nanoflakes for toxic gas detection. Nanotechnol 2017;28(43):435502. [14] Jha RK, Wan M, Jacob C, Guha PK. Ammonia vapour sensing properties of in situ polymerized conducting PANI-nanofiber/WS2 nanosheet composites. New J Chem 2018;42(1):735e45. [15] Ghosh R, Nayak A, Santra S, Pradhan D, Guha PK. Enhanced ammonia sensing at room temperature with reduced graphene oxide/tin oxide hybrid film. RSC Adv 2015;5:50165e73. Integrated CMOS-based sensors for gas and odor detection 487

[16] Ghosh R, Santra S, Ray SK, Guha PK. Pt-functionalized reduced graphene oxide for excellent hydrogen sensing at room temperature. Applied Physics Letters 2015;107. 153102, 5pp. [17] Nayak AK, Ghosh R, Santra S, Guha PK, Pradhan D. Hierarchical nanostructured WO3eSnO2 for selective sensing of volatile organic compounds. Nanoscale 2015;7: 12460e73. [18] Santra S, Guha PK, Ali SZ, Hiralal P, Unalan HE, Covington JA, Amaratunga GAJ, Milne WI, Gardner JW, Udrea F. ZnO nanowires grown on SOI CMOS substrate for ethanol sensing. Sensor Actuator B 2010b;146:559e65. [19] Santra S, Ali SZ, Guha PK, Covington JA, Zhong G, Robertson J, Milne WI, Gardner JW, Udrea F. Post-CMOS wafer level growth of carbon nanotubes for low-cost microsensorsda proof of concept. Nanotechnol 2010a;21:485301. 7pp. [20] Bota SA, Dieguez A, Merino JL, Casanova R, Samitier J, Cane C. A monolithic interface circuit for gas sensor arrays: control and measurement. Analog Integr Circuits Signal Process 2004;40:175e84. [21] Barrettino D, Graf M, Haﬁzovic S, Taschini S, Hagleitner C, Hierlemann A, Baltes H. A single-chip CMOS micro-hotplate array for hazardous-gas detection and material characterization. In: International solid-state circuit conference. San Francisco, CA; 2004. [22] De Marcellis A, Depari A, Ferri G, Flammini A, Marioli D, Stornelli V, Taroni A. Un- calibrated integrable wide-range single-supply portable interface for resistance and parasitic capacitance determination. Sensor Actuator B Chem 2008;132. 477e184. [23] Grassi M, Malcovati P, Baschirotto A. Wide-range integrated gas sensor interface based on a resistance-to-number converter technique with the oscillator decoupled from the input device. In: Proceedings of IEEE symposium on circuits and systems. Kos Island, Greece; 2006. [24] Grassi M, Malcovati P, Baschirotto A. A 141-dB dynamic range CMOS gas-sensor interface circuit without calibration with 16-bit digital output word. IEEE J Solid State Circuits 2007;42:1543e54. [25] Barrettino D, Graf M, Zimmermann M, Hierlemann A, Baltes H, Hahn S, Barsan N, Weimar U. A smart single-chip micro-hotplate-based chemical sensor system in CMOS-technology. Int Symp Circuits Syst 2002;2:157e60. [26] Koickal TJ, Hamilton A, Tan SL, Covington J, Gardner JW, Pearce T. Smart interface circuit to ameliorate loss of measurement range in chemical microsensor arrays. In: Pro- ceedings of the IEEE instrumentation and measurement technology conference. Ottawa, Ont., Canada; 2005. [27] Koickal TJ, Hamilton A, Tan SL, Covington JA, Gardner JW, Pearce TC. Analog VLSI circuit implementation of an adaptive neuromorphic olfaction chip. IEEE Trans Circuits Syst I: Regular Paper 2007;54(1):60e73. [28] Gardner JW, Bartlett PN. Electronic noses: principles and application. Oxford: Oxford Uni- versity Press; 1999. [29] Pearce T, Schiffman SS, Nagle HT, Gardner JW. Handbook of machine olfaction: electronic nose technology. Weinheim: Wiley VCH; 2003. [30] Iwaki T, Covington JA, Gardner JW. Identiﬁcation of different vapors using a single temperature modulated polymer sensor with a novel signal processing technique. IEEE Sens J 2009;9:314e28. This page intentionally left blank Index

Note: ‘Page numbers followed by “f ” indicate ﬁgures, “t” indicates tables’.

A synthesis of “Ablation plume”, 164 arc discharge, 192 Acceptor-doped materials, 366e367 chemical vapor deposition Adsorption (CVD), 193 molecular mechanism of, 261e265 laser ablation, 192e193 phenomena, 250e261 18650 cells, 190e191 processes, optical probing of, 245e246 Charge carrier transport, 104e119 Air constituents, competitive adsorption Chemical vapor deposition (CVD), of, 253e259 164e165, 193, 474 Ammonia, 310e311 Chemoresistive gas sensors, 39e40 Analog-to-digital converter (ADC), Clean semiconductor surface, 117e118 479e480 CNT/polymer nanocomposite sensing Arc discharge, 192 materials, 209e210 on fabric substrate, 215e216, 215f B on textile substrate, 212e215 e Base air resistance, 18 21 CO2, 329e331, 331f Binary switch behavior, 323 Commercial ﬁnite element method Breathing mode, 387 program, 362 BTEX, 223 Commercial sensor systems, 310e311 compounds, 223e225, 224t Competitive adsorption of air constituents, 253e259 C Complementary metal oxide Calixarenes, 229e230 semiconductor (CMOS) Cambridge CMOS Sensor (CCS), 454 analog-to-digital converter (ADC), Carbon nanomaterials, 226 479e480 Carbon nanotube chemical vapor deposition (CVD), 474 characterization of future trends, 483e485 Raman spectroscopy, 200e201, 200f integrated multisensor and sensor array scanning electron microscope (SEM), systems, 480e483, 481f, 483f 203e204, 203fe204f interface circuitry, 475e480 transmission electron microscopy driving circuits for microheaters, 476 (TEM), 204e205, 205f hybrid approach, 475 X-ray diffraction, 201e202, 202f monolithic approach, 475 preparation of sensing material interface circuit, 478 dip-coating, 198 temperature control unit, 476e477 drop-coating, 197 low-cost semiconductor gas microsensors, electron beam (E-beam) evaporation, 466e467 198e199 microcalorimetric complementary metal screen-printing, 197e198 oxide semiconductor gas sensor, spin-coating, 194e196 469e472 sensing mechanism of, microheaters, 469 205e206

489 j 490 Index

Complementary metal oxide semiconductor Dielectric material-gate field-effect (CMOS) (Continued) transistor, 31 microresistive complementary metal Dielectric thin films, 439e441 oxide semiconductor gas sensors, Dilute limit, 118 467e469 Diode-type sensors, 33e34 new-generation microresistive Dip-coating, 198 sensors, 468 Dipole formation, interfacial layer, sensing materials, 472e475 119e120 silicon on insulator (SOI), 468 Direct and indirect thermoelectric gas Taguchi gas sensors (TGS), sensors, 350e352, 351f 467e468, 467f Donor-doped materials, 366e367 techniques, 474e475 Drift velocity, 76 volatile organic compounds (VOCs), 472 Drop-coating technique, 197, 211, 211f Conduction electrons, 74 Dynamic sensor operation, 332e334 Conduction mechanism in SMOXs gas breathing mode, 387 sensors electrical multiparameter readout n-type SnO2,62e67 (EMR), 386 p-type CuO, 58e62 gas-sensitive field-effect transistors sample preparation and experimental (GasFETs), 388, 398e404 conditions, 57e58 combined methods, 403e404 Conductive polymers, 81, 81f current compensation mode, 403 Conductometric sensor, 169e170, 348 gate biasecycled operation, Contact potential differences (CPDs), 57, 401e403, 402f 58f, 62f, 64f, 97 SiC-FET sensors, 399e400 Conventional pellistors, 469e472 temperature-cycled operation, CPDs. See Contact potential differences 399e400 (CPDs) metal oxide semiconductor (MOS) gas Crocheting technique, 206e207 sensors, 385, 388e398 CVD. See Chemical vapor deposition field effect, 396e397, 398f (CVD) optical excitation, 398 Cyclodextrins (CD), 226e229, 228t temperature-cycled operation, 390e396, 392f, 394fe395f D stability problems, 386 Dangling bonds, 99e101 systematic data evaluation, 386e387 Deep cavitands, 230e232 temperature-cycled operation Deep reactiveeion-etched (DRIE) (TCO), 388 membrane, 436e437 transient response, 387 Depletion approximation, 105e106 Depletion layer model, 56 E Deposition techniques, 162 Electrical gas sensors, 272 one-dimensional nanostructures, Electrical multiparameter readout 166e169, 167f (EMR), 386 two-dimensional nanostructures, Electric field and capacitance, 104e109 163e166 Electrode configuration, 85e90 Derivatives, 229e230 Electrode geometry, 90e95 Device integration, 170 Electrode materials, influence of, Device structure, 9e10, 9f 77e85 Index 491

Electrode-oxide interfaces, gas sensor F operation Fabrication, 10 charge carrier transport in, 104e119 Fabric-based embroidered gas sensors, electrode configuration, 85e90 212e213 electrode geometry, 90e95 Fabric-based screen-printed gas sensors, electrode-oxide semiconductor interfaces 213e215, 214f contacts with surface states and Field-effect transistor-type gas sensors, interfacial layer, 99e103 27e31 ideal contact of metal and oxide Field emission (FE), 113e114 semiconductor, 95e99 Flame spray pyrolysis (FSP), 431, 432f image force effects on the barrier Focused ion beam scanning electron height, 103e104 microscope (FIB-SEM), 445e446 gas/solid interactions in dipole formation in interfacial layer, G 119e120 Gas response, disturbances to, 13 hydrogen adsorption in Schottky Gas sensing characteristics, 11e13 barrier junction, 120e122 Gas-sensitive devices, 414e416 Schottky barrier junction, Gas-sensitive field-effect transistors 122e124 (GasFETs), 388, 398e404 influence of electrode materials, 77e85 combined methods, 403e404 metals and conduction, 74e76 current compensation mode, 403 semiconductor gas sensor, 72e73 gate biasecycled operation, Taguchi sensor, 72 401e403, 402f Electrode-oxide semiconductor interfaces SiC-FET sensors, 399e400 contacts with surface states and interfacial temperature-cycled operation, 399e400 layer, 99e103 Gas sensor operation, electrode-oxide ideal contact of metal and oxide interfaces in semiconductor, 95e99 charge carrier transport in, 104e119 image force effects on the barrier height, electrode configuration, 85e90 103e104 electrode geometry, 90e95 Electrodes and CNT/polymer electrode-oxide semiconductor interfaces nanocomposites for textile-based contacts with surface states and sensors, 206e210 interfacial layer, 99e103 Electron affinity, 46e47 ideal contact of metal and oxide Electron beam evaporation (EBE), semiconductor, 95e99 163, 198e199 image force effects on the barrier Electron energy band diagram, 102f height, 103e104 Embroidery technique, 207 gas/solid interactions in English Electric Valve (EEV) Ltd, dipole formation in interfacial layer, 469e472 119e120 E-textiles, 189, 191 hydrogen adsorption in Schottky 3+ e Eu :ZrO2 barrier junction, 120 122 characterization, 289e290, 290f Schottky barrier junction, 122e124 oxygen sensing, 290e291, 292f influence of electrode materials, 77e85 preparation, 289e290, 290f metals and conduction, 74e76 sensing mechanism, 291e294, semiconductor gas sensor, 72e73 293fe294f Taguchi sensor, 72 492 Index

Gas sensors, 32e34, 186e187 Ionic direct thermoelectric gas sensors, Gas/solid interactions in 378e379 dipole formation in interfacial layer, Isothermal PoissoneBoltzmann equation, 119e120 362 hydrogen adsorption in Schottky barrier junction, 120e122 L Schottky barrier junction, 122e124 Light-addressable potentiometric sensor Gold, 83 (LAPS), 144e147 Limit of detection (LOD), 225 H Liquid phase growth methods, 168e169 e H2O adsorbates, 259e261 Llaser ablation, 192 193 Homogeneous semiconductors, 365 Low-cost semiconductor gas microsensors, “Hot wire” type, 9e10 466e467 fi H2O vapor, response to, 248e249 Low temperature co red ceramic “Hybrid ” chemical gas densor, 187 (LTCC), 312 Hydrogen adsorption in Schottky barrier junction, 120e122 M Hydrogen-containing gases, Macrocyclic compounds, 225e226 320e324, 322f Magnesium oxide (MgO), 327e328 Hydrogen detection, 323e324 Metal-insulatoresemiconductor capacitor type sensors, 32e33, 33f I Metal insulator semiconductor (MIS) gas Ideal contact of metal, 95e99 sensor, 310e311 Ideal gas sensors, 134 Metal oxides (MOXs), 161e162, III-nitrides, 243 200e206, 239e240, 272 Image force effects on barrier height, Metal oxide semiconductor capacitor, 103e104 142e144 Immersion-coating technique, 210, 211f Metal oxide semiconductor field-effect Impurity stabilized, 118 transistor (MOSFETs), 148e150 Inflammable gases, response to, 22e23 Metal oxide semiconductor (MOS) gas InGaN/GaN nanowire heterostructure sensors, 151, 385, 388e398 arrays, 239e242 band diagram modulation, 153e154 Inhomogeneous semiconductors, 365 field effect, 396e397, 398f Integrated multisensor and sensor array optical excitation, 398 e systems, 480 483, 481f, 483f SnO2 bands, 151e152 Interdiffusion, 118e119 temperature-cycled operation, 390e396, Interface circuitry, 475e480 392f, 394fe395f driving circuits for microheaters, 476 Metal oxide sensing films, 194e199 hybrid approach, 475 Metals and conduction, 74e76 monolithic approach, 475 “Metalesemiconductor contact diode”, sensing material interface circuit, 478 33e34 temperature control unit, 476e477 Metal-semiconductor interface, 104e109 Interfacial layer, 99e103 Metal-semiconductor interfacial zones, structure of, 115e119 116f Internal sensitivities, 137 Microcalorimetric complementary metal Internet of Things (IoT), 134, 455 oxide semiconductor gas sensor, Intrinsic emission, 279 469e472 Index 493

Microelectromechanical system (MEMS) Molecular mechanism of adsorption, sensor, 9e10, 35 261e265 Microheaters, 469 Monolayer formation, 118 Micromachined semiconductor gas sensors Monolayers, 118e119 applications, 454e458 Mott barrier, 97e98 CMOS-compatible process, 439 MOX gas sensors, 189 complementary metal oxide MOX materials, 187 semiconductor (CMOS), 414, Multivariable sensing, 286e287 437e441 concept and technologies, 416e425 N gas-sensitive devices, 414e416 Nanocrystalline form, 274 polysilicon heaters, 415 Nanostructured gas sensing layers, temperature modulation, 415 445e450, 445f, 448fe449f manufacturing, 454e458 Nanowires, 448e449 microelectromechanical systems (MEMS) New-generation microresistive sensors, devices, 413e414 468 microhotplate performance, 418e421, Nitrogen oxides, 310e311 419fe420f Nonhydrogen-containing gases, microhotplate realization, 418e421, 324e327, 325f 419fe420f Nonresistive sensors, 27 e microhotplate reliability, 421 425 NOx, 329e331, 331f micromachined field-effect gas sensors, n-type, 47e56 e e 442 444, 443f 444f n-type SnO2,62e67 micromachined metal oxide gas sensors, 425e437 O flame spray pyrolysis (FSP), 431, 432f One-dimensional approach, 52 packaging, 435e437, 436f One-dimensional nanostructures, temperature modulation, 166e169, 167f 432e435, 434f One-electrode configuration, 87f thick gas-sensitive films, Operating temperature, 12e13 428e431, 429f Optical probing of adsorption processes, thin gas-sensitive films, 425e428, 427f 245e246 MOSFET switch, 439 Ordinary least square (OLS), 287 nanostructured gas sensing layers, Organic precursor, 278 445e450, 445f, 448fe449f Output characteristics, MOSFET, 149 polymeric foil, 450e454 Oxide semiconductor, 95e99 printing semiconductor gas sensors, “Oxide semiconductor gas sensors”,6 452e454 Oxide semiconductor-gate field-effect semiconductor gas sensors, transistor, 29e31 450e451, 452f Oxide-semiconductor interface, tunneling products, 454e458 effects in, 112e115 robust high-temperature tungsten-based Oxidizing gases, response to, 23e25, SOI microhotplates, 441 246e248 thermal design, 416e418, 417f Oxygen, 277, 328e329 Microresistive complementary metal concentration cell type sensors, 31e32 oxide semiconductor gas sensors, nitrogen mixtures, 374e375, 374f 467e469 response to, 18e21 “Mixed potential” type sensors, 31e32 494 Index

P Resistor-type sensors Palladiumegold, 85 extensions, 25e27 Palladiumesilver, 84 field-effect transistor-type gas sensors Petrochemical industry, 223e224 dielectric material-gate field-effect Photoluminescence-based gas sensors transistor, 31 collisional energy transfer process, 273 oxide semiconductor-gate field-effect conductometric sensing, 272f transistor, 29e31 conventional chemiresistive gas principle, 27e28 sensing, 272 solid electrolyte-gate field-effect fluorescence, 272 transistor, 28e29 nanocrystalline form, 274 gas sensing characteristics, 11e13 oxygen sensing mechanism, 277 gas sensors, 32e34 progress, 275e277 nonresistive sensors, 27 rare eartheactivated inorganic sensor oxygen concentration cell type sensors, materials, 273e275 31e32 SterneVolmer relationship, 273 receptor function and transducer Photoluminescent InGaN/GaN nanowire function, 14e18 arrays, 243e245 response to inflammable gases, 22e23 Platinum, 84 response to oxidizing gases, 23e25 Platinumegold, 84e85 response to oxygen, 18e21 Platinumesilver, 84 semiconductor oxygen sensors, 13e14 PL response, 246e250 sensing materials and devices, 6e10 dependence of, 250e253 Response behavior, 246 Poisson’s equation, 52 Response transients, 11e12, 11f Polyimide (PI), 450e451 Polymeric foil, 450e454 S printing semiconductor gas sensors, Scanning electron microscope (SEM), 452e454 203e204, 203fe204f semiconductor gas sensors, 450e451, fabric-based embroidered gas sensors, 452f 212e213 Polysilicon heaters, 415 fabric-based screen-printed gas sensors, “Precursors”, 164e165 213e215, 214f 3+ e Pr :(K0.5Na0.5)NbO3 Schottky barrier junction, 97 98, 110f, oxygen sensing, 298e299 122e124 potassiumesodium niobate, 298 hydrogen adsorption in, 120e122 synthesis, 298 Screen-printing technology, 41e42, p-type, 47e56 197e198, 208e209 CuO, 58e62 SEM. See Scanning electron microscope (SEM) Q Semiconducting direct thermoelectric gas Quartz crystal microbalance (QMB), 327 sensors acceptor-doped materials, 366e367 R commercial finite element method Raman scattering analyses, 289e290 program, 362 Raman spectroscopy, 200e201, 200f conductometric gas sensors, 348 Ratiometric sensor material, 277 definition, 353e379 Receptor function, 14e18, 40e41 direct and indirect thermoelectric gas “Reduced resistance”,17e18 sensors, 350e352, 351f Reducing gases, response to, 249e250 donor-doped materials, 366e367 Index 495

future trends, 380e381 resistor-type sensors homogeneous semiconductors, 365 extensions, 25e27 inhomogeneous semiconductors, 365 field-effect transistor-type gas sensors, ionic direct thermoelectric gas sensors, 27e31 378e379 gas sensing characteristics, 11e13 isothermal PoissoneBoltzmann gas sensors, 32e34 equation, 362 nonresistive sensors, 27 measurements, 367e378 oxygen concentration cell type sensors, measurement techniques, 353e357, 31e32 355fe356f receptor function and transducer modeling and simulation, 357e367, function, 14e18 358f, 360f response to inflammable gases, 22e23 motivation for research, 347e349 response to oxidizing gases, 23e25 oxygen/nitrogen mixtures, 374e375, response to oxygen, 18e21 374f semiconductor oxygen sensors, 13e14 propane concentration profile, 371 sensing materials and devices, 6e10 research activities, 353 sensor blocks, 135 thermoelectric power, 349e350 evaluation of resolution, 139e142 Wolkenstein isotherm, 365e366 resolution, 138e139 yttria-stabilized zirconia (YSZ), 348 response curve, 136 Semiconducting metal oxide (SMOXs) sensitivity, 137e138 gas sensors, 41e47, 41f solid-state sensors, 134 conduction mechanism in Semiconductor oxygen sensors, 13e14 n-type SnO2,62e67 Semiconductor technology, 243, 273 p-type CuO, 58e62 Sensing materials, 472e475 sample preparation and experimental devices, 6e10 conditions, 57e58 Sensing materials mechanism n-type, 47e56 carbon nanotube, 205e206 p-type, 47e56 Sensirion Gas sensor Platform (SGP), Semiconducting metal oxides (SMOXs), 455e457, 456f 39e40 Sensitizers, 8 Semiconductor gas sensors, 4, 72e73, Sensor assembly, 199e200 239e242 for textile-based gas sensors classification of, 5e6, 5f drop-coating technique, 211, 211f ideal gas sensors, 134 immersion-coating technique, industrial processes and pollution control, 210, 211f 133 Sensor blocks, 135 light-addressable potentiometric sensor, evaluation of resolution, 139e142 144e147 resolution, 138e139 metal oxide semiconductor capacitor, response curve, 136 142e144 sensitivity, 137e138 metal oxide semiconductor field-effect Sensor packaging, 335e336, 335f transistor (MOSFETs), 148e150 “Short-term effect of water vapor”,13 metal oxide semiconductors Silicon carbide field effect gas sensors (MOS), 151 advanced data evaluation, 332e334 band diagram modulation, 153e154 ammonia, 310e311 SnO2 bands, 151e152 applications 496 Index

Silicon carbide field effect gas sensors Solid-state sensors, 134 (Continued) Spin-coating, 194e196 field tests, 336e338, 337f SterneVolmer law, 280 sensor packaging, 335e336, 335f Surface states, 99e103 commercial sensor systems, 310e311 Synthesis of carbon nanotube dynamic sensor operation, 332e334 arc discharge, 192 field effect devices, 311 chemical vapor deposition (CVD), 193 low temperature cofired ceramic laser ablation, 192e193 (LTCC), 312 Systematic data evaluation, 386e387 metal insulator semiconductor (MIS) gas sensor, 310e311 T nitrogen oxides, 310e311 Taguchi gas sensors (TGS), 467e468, 467f sensing mechanisms Taguchi sensor, 72 3+ binary switch behavior, 323 Tb :CePO4 CO2, 329e331, 331f characterization, 295, 296f hydrogen-containing gases, gas sensing, 295e297, 296f 320e324, 322f orthophosphates, 294 hydrogen detection, 323e324 preparation, 295, 296f magnesium oxide (MgO), 327e328 X-ray photoelectron spectroscopy new material combinations, 327e328 (XPS), 297 nonhydrogen-containing gases, TEM. See Transmission electron 324e327, 325f microscopy (TEM) NOx, 329e331, 331f Temperature-cycled operation oxygen, 328e329 (TCO), 388 quartz crystal microbalance Temperature modulation, 415, (QMB), 327 432e435, 434f tailor-made sensing layers, Template-assisted methods, 169 328e329, 328f Textile-based electrode, preparation of ultrahigh vacuum (UHV), 323 crocheting technique, 206e207 voltage drops, 318e319 embroidery technique, 207 transduction, 312e327 screen printing technique, 208e209 transducer platform, 313e316, 314f Textile-based gas sensors, sensor transduction mechanisms, assembly for 316e318, 318f drop-coating technique, 211, 211f Silicon on insulator (SOI), 468 immersion-coating technique, 210, 211f Silver, 83 Thermal design, 416e418, 417f Single-crystalline, 295 Thermal evaporation (TE) technique, 163 SMOX-based sensors, 40e41 Thermionic field emission (TFE), 3+ Sm :TiO2 113e114 characterization, 278e280 Thermoelectric power, 349e350 mathematical model, 282e285, 283f Thick gas-sensitive films, 428e431, 429f multivariable sensing, 286e287, 286f Thin gas-sensitive films, 425e428, 427f PL-based oxygen sensing, 280e282, 281f Transducer function, 14e18 preparation, 278e280 Transduction principles and related novel sensing mechanism, 282e285 devices, 170e174 Solid electrolyte-gate field-effect Transient response, 387 transistor, 28e29 Transistor outline (TO) headers, 435 Index 497

Transmission electron microscopy (TEM), W 204e205, 205f “Wet” process, 10 Transport mechanism across the junction Wolkenstein isotherm, 365e366 barrier, 109e112 Tunneling effects, in oxide- X semiconductor interface, 112e115 X-ray diffraction (XRD), 201e202, Two-dimensional nanostructures, 202f, 289e290 163e166 X-ray photoelectron spectroscopy (XPS), 297 U Ultrahigh vacuum (UHV), 323 Y “Utility factor”,12e13 Yttria-stabilized zirconia (YSZ), 348 V Z

Vapor phase growth methods, Zirconium dioxide (ZrO2), 288e294 167e168, 446 ZnO nanowires, 451 Volatile organic compounds (VOCs), 223, 472 Voltage drops, 318e319 This page intentionally left blank